security: Enhance Hugging Face model loading security - resolves #136

BREAKING CHANGE: trust_remote_code now defaults to False for security

- Set trust_remote_code=False by default in HFChat class
- Add explicit trust_remote_code parameter to HFChat.__init__()
- Add security warning when trust_remote_code=True is used
- Update get_llm() function to support trust_remote_code parameter
- Update benchmark utilities (load_hf_model, load_vllm_model, load_qwen_vl_model)
- Add comprehensive documentation for security implications

Security Benefits:
- Prevents arbitrary code execution from compromised model repositories
- Requires explicit opt-in for models that need remote code execution
- Shows clear warnings when security is reduced
- Follows security-by-default principle

Migration Guide:
- Most users: No changes needed (more secure by default)
- Users with models requiring remote code: Add trust_remote_code=True explicitly
- Config users: Add 'trust_remote_code': true to LLM config if needed

Fixes #136

.github/ISSUE_TEMPLATE/bug_report.yml

@@ -0,0 +1,50 @@
+name: Bug Report
+description: Report a bug in LEANN
+labels: ["bug"]
+
+body:
+  - type: textarea
+    id: description
+    attributes:
+      label: What happened?
+      description: A clear description of the bug
+    validations:
+      required: true
+
+  - type: textarea
+    id: reproduce
+    attributes:
+      label: How to reproduce
+      placeholder: |
+        1. Install with...
+        2. Run command...
+        3. See error
+    validations:
+      required: true
+
+  - type: textarea
+    id: error
+    attributes:
+      label: Error message
+      description: Paste any error messages
+      render: shell
+
+  - type: input
+    id: version
+    attributes:
+      label: LEANN Version
+      placeholder: "0.1.0"
+    validations:
+      required: true
+
+  - type: dropdown
+    id: os
+    attributes:
+      label: Operating System
+      options:
+        - macOS
+        - Linux
+        - Windows
+        - Docker
+    validations:
+      required: true

.github/ISSUE_TEMPLATE/config.yml

@@ -0,0 +1,8 @@
+blank_issues_enabled: true
+contact_links:
+  - name: Documentation
+    url: https://github.com/LEANN-RAG/LEANN-RAG/tree/main/docs
+    about: Read the docs first
+  - name: Discussions
+    url: https://github.com/LEANN-RAG/LEANN-RAG/discussions
+    about: Ask questions and share ideas

.github/ISSUE_TEMPLATE/feature_request.yml

@@ -0,0 +1,27 @@
+name: Feature Request
+description: Suggest a new feature for LEANN
+labels: ["enhancement"]
+
+body:
+  - type: textarea
+    id: problem
+    attributes:
+      label: What problem does this solve?
+      description: Describe the problem or need
+    validations:
+      required: true
+
+  - type: textarea
+    id: solution
+    attributes:
+      label: Proposed solution
+      description: How would you like this to work?
+    validations:
+      required: true
+
+  - type: textarea
+    id: example
+    attributes:
+      label: Example usage
+      description: Show how the API might look
+      render: python

.github/pull_request_template.md

@@ -0,0 +1,13 @@
+## What does this PR do?
+
+<!-- Brief description of your changes -->
+
+## Related Issues
+
+Fixes #
+
+## Checklist
+
+- [ ] Tests pass (`uv run pytest`)
+- [ ] Code formatted (`ruff format` and `ruff check`)
+- [ ] Pre-commit hooks pass (`pre-commit run --all-files`)

.github/workflows/build-reusable.yml

@@ -17,26 +17,17 @@ jobs:
       - uses: actions/checkout@v4
         with:
           ref: ${{ inputs.ref }}
+          submodules: recursive
 
-      - name: Setup Python
-        uses: actions/setup-python@v5
+      - name: Install uv and Python
+        uses: astral-sh/setup-uv@v6
         with:
           python-version: '3.11'
 
-      - name: Install uv
-        uses: astral-sh/setup-uv@v4
-
-      - name: Install ruff
+      - name: Run pre-commit with only lint group (no project deps)
         run: |
-          uv tool install ruff
+          uv run --only-group lint pre-commit run --all-files --show-diff-on-failure
 
-      - name: Run ruff check
-        run: |
-          ruff check .
-
-      - name: Run ruff format check
-        run: |
-          ruff format --check .
 
   build:
     needs: lint
@@ -54,6 +45,17 @@ jobs:
             python: '3.12'
           - os: ubuntu-22.04
             python: '3.13'
+          # ARM64 Linux builds
+          - os: ubuntu-24.04-arm
+            python: '3.9'
+          - os: ubuntu-24.04-arm
+            python: '3.10'
+          - os: ubuntu-24.04-arm
+            python: '3.11'
+          - os: ubuntu-24.04-arm
+            python: '3.12'
+          - os: ubuntu-24.04-arm
+            python: '3.13'
           - os: macos-14
             python: '3.9'
           - os: macos-14
@@ -92,14 +94,11 @@ jobs:
           ref: ${{ inputs.ref }}
           submodules: recursive
 
-      - name: Setup Python
-        uses: actions/setup-python@v5
+      - name: Install uv and Python
+        uses: astral-sh/setup-uv@v6
         with:
           python-version: ${{ matrix.python }}
 
-      - name: Install uv
-        uses: astral-sh/setup-uv@v6
-
       - name: Install system dependencies (Ubuntu)
         if: runner.os == 'Linux'
         run: |
@@ -108,13 +107,46 @@ jobs:
             pkg-config libabsl-dev libaio-dev libprotobuf-dev \
             patchelf
 
-          # Install Intel MKL for DiskANN
+          # Debug: Show system information
+          echo "🔍 System Information:"
+          echo "Architecture: $(uname -m)"
+          echo "OS: $(uname -a)"
+          echo "CPU info: $(lscpu | head -5)"
+
+          # Install math library based on architecture
+          ARCH=$(uname -m)
+          echo "🔍 Setting up math library for architecture: $ARCH"
+
+          if [[ "$ARCH" == "x86_64" ]]; then
+            # Install Intel MKL for DiskANN on x86_64
+            echo "📦 Installing Intel MKL for x86_64..."
             wget -q https://registrationcenter-download.intel.com/akdlm/IRC_NAS/79153e0f-74d7-45af-b8c2-258941adf58a/intel-onemkl-2025.0.0.940.sh
             sudo sh intel-onemkl-2025.0.0.940.sh -a --components intel.oneapi.lin.mkl.devel --action install --eula accept -s
             source /opt/intel/oneapi/setvars.sh
             echo "MKLROOT=/opt/intel/oneapi/mkl/latest" >> $GITHUB_ENV
             echo "LD_LIBRARY_PATH=/opt/intel/oneapi/compiler/latest/linux/compiler/lib/intel64_lin" >> $GITHUB_ENV
             echo "LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/opt/intel/oneapi/mkl/latest/lib/intel64" >> $GITHUB_ENV
+            echo "✅ Intel MKL installed for x86_64"
+
+            # Debug: Check MKL installation
+            echo "🔍 MKL Installation Check:"
+            ls -la /opt/intel/oneapi/mkl/latest/ || echo "MKL directory not found"
+            ls -la /opt/intel/oneapi/mkl/latest/lib/ || echo "MKL lib directory not found"
+
+          elif [[ "$ARCH" == "aarch64" ]]; then
+            # Use OpenBLAS for ARM64 (MKL installer not compatible with ARM64)
+            echo "📦 Installing OpenBLAS for ARM64..."
+            sudo apt-get install -y libopenblas-dev liblapack-dev liblapacke-dev
+            echo "✅ OpenBLAS installed for ARM64"
+
+            # Debug: Check OpenBLAS installation
+            echo "🔍 OpenBLAS Installation Check:"
+            dpkg -l | grep openblas || echo "OpenBLAS package not found"
+            ls -la /usr/lib/aarch64-linux-gnu/openblas/ || echo "OpenBLAS directory not found"
+          fi
+
+          # Debug: Show final library paths
+          echo "🔍 Final LD_LIBRARY_PATH: $LD_LIBRARY_PATH"
 
       - name: Install system dependencies (macOS)
         if: runner.os == 'macOS'
@@ -124,11 +156,24 @@ jobs:
 
       - name: Install build dependencies
         run: |
-          uv pip install --system scikit-build-core numpy swig Cython pybind11
-          if [[ "$RUNNER_OS" == "Linux" ]]; then
-            uv pip install --system auditwheel
+          uv python install ${{ matrix.python }}
+          uv venv --python ${{ matrix.python }} .uv-build
+          if [[ "$RUNNER_OS" == "Windows" ]]; then
+            BUILD_PY=".uv-build\\Scripts\\python.exe"
           else
-            uv pip install --system delocate
+            BUILD_PY=".uv-build/bin/python"
+          fi
+          uv pip install --python "$BUILD_PY" scikit-build-core numpy swig Cython pybind11
+          if [[ "$RUNNER_OS" == "Linux" ]]; then
+            uv pip install --python "$BUILD_PY" auditwheel
+          else
+            uv pip install --python "$BUILD_PY" delocate
+          fi
+
+          if [[ "$RUNNER_OS" == "Windows" ]]; then
+            echo "$(pwd)\\.uv-build\\Scripts" >> $GITHUB_PATH
+          else
+            echo "$(pwd)/.uv-build/bin" >> $GITHUB_PATH
           fi
 
       - name: Set macOS environment variables
@@ -264,18 +309,66 @@ jobs:
 
       - name: Install built packages for testing
         run: |
-          # Create a virtual environment with the correct Python version
+          # Create uv-managed virtual environment with the requested interpreter
+          uv python install ${{ matrix.python }}
           uv venv --python ${{ matrix.python }}
           source .venv/bin/activate || source .venv/Scripts/activate
 
-          # Install packages using --find-links to prioritize local builds
-          uv pip install --find-links packages/leann-core/dist --find-links packages/leann-backend-hnsw/dist --find-links packages/leann-backend-diskann/dist packages/leann-core/dist/*.whl || uv pip install --find-links packages/leann-core/dist packages/leann-core/dist/*.tar.gz
-          uv pip install --find-links packages/leann-core/dist packages/leann-backend-hnsw/dist/*.whl
-          uv pip install --find-links packages/leann-core/dist packages/leann-backend-diskann/dist/*.whl
-          uv pip install packages/leann/dist/*.whl || uv pip install packages/leann/dist/*.tar.gz
+          if [[ "$RUNNER_OS" == "Windows" ]]; then
+            UV_PY=".venv\\Scripts\\python.exe"
+          else
+            UV_PY=".venv/bin/python"
+          fi
 
-          # Install test dependencies using extras
-          uv pip install -e ".[test]"
+          # Install test dependency group only (avoids reinstalling project package)
+          uv pip install --python "$UV_PY" --group test
+
+          # Install core wheel built in this job
+          CORE_WHL=$(find packages/leann-core/dist -maxdepth 1 -name "*.whl" -print -quit)
+          if [[ -n "$CORE_WHL" ]]; then
+            uv pip install --python "$UV_PY" "$CORE_WHL"
+          else
+            uv pip install --python "$UV_PY" packages/leann-core/dist/*.tar.gz
+          fi
+
+          PY_TAG=$($UV_PY -c "import sys; print(f'cp{sys.version_info[0]}{sys.version_info[1]}')")
+
+          if [[ "$RUNNER_OS" == "macOS" ]]; then
+            if [[ "${{ matrix.os }}" == "macos-13" ]]; then
+              export MACOSX_DEPLOYMENT_TARGET=13.3
+            elif [[ "${{ matrix.os }}" == "macos-14" ]]; then
+              export MACOSX_DEPLOYMENT_TARGET=14.0
+            elif [[ "${{ matrix.os }}" == "macos-15" ]]; then
+              export MACOSX_DEPLOYMENT_TARGET=15.0
+            fi
+          fi
+
+          HNSW_WHL=$(find packages/leann-backend-hnsw/dist -maxdepth 1 -name "*-${PY_TAG}-*.whl" -print -quit)
+          if [[ -z "$HNSW_WHL" ]]; then
+            HNSW_WHL=$(find packages/leann-backend-hnsw/dist -maxdepth 1 -name "*-py3-*.whl" -print -quit)
+          fi
+          if [[ -n "$HNSW_WHL" ]]; then
+            uv pip install --python "$UV_PY" "$HNSW_WHL"
+          else
+            uv pip install --python "$UV_PY" ./packages/leann-backend-hnsw
+          fi
+
+          DISKANN_WHL=$(find packages/leann-backend-diskann/dist -maxdepth 1 -name "*-${PY_TAG}-*.whl" -print -quit)
+          if [[ -z "$DISKANN_WHL" ]]; then
+            DISKANN_WHL=$(find packages/leann-backend-diskann/dist -maxdepth 1 -name "*-py3-*.whl" -print -quit)
+          fi
+          if [[ -n "$DISKANN_WHL" ]]; then
+            uv pip install --python "$UV_PY" "$DISKANN_WHL"
+          else
+            uv pip install --python "$UV_PY" ./packages/leann-backend-diskann
+          fi
+
+          LEANN_WHL=$(find packages/leann/dist -maxdepth 1 -name "*.whl" -print -quit)
+          if [[ -n "$LEANN_WHL" ]]; then
+            uv pip install --python "$UV_PY" "$LEANN_WHL"
+          else
+            uv pip install --python "$UV_PY" packages/leann/dist/*.tar.gz
+          fi
 
       - name: Run tests with pytest
         env:

.gitignore

@@ -18,10 +18,12 @@ demo/experiment_results/**/*.json
 *.eml
 *.emlx
 *.json
+*.png
 !.vscode/*.json
 *.sh
 *.txt
 !CMakeLists.txt
+!llms.txt
 latency_breakdown*.json
 experiment_results/eval_results/diskann/*.json
 aws/
@@ -93,18 +95,13 @@ packages/leann-backend-diskann/third_party/DiskANN/_deps/
 batchtest.py
 tests/__pytest_cache__/
 tests/__pycache__/
-paru-bin/
-
-CLAUDE.md
-CLAUDE.local.md
-.claude/*.local.*
-.claude/local/*
-
 benchmarks/data/
-!benchmarks/data/prompts_g5/*.txt
-!benchmarks/run_all.sh
-!benchmarks/run_speed_bench_all.sh
-!benchmarks/simple_mac_tpt_test.py
-!benchmarks/run_speed_bench_all.sh
-!benchmarks/run_speed_bench_all.sh
-!benchmarks/run_speed_bench_all.sh
+
+## multi vector
+apps/multimodal/vision-based-pdf-multi-vector/multi-vector-colpali-native-weaviate.py
+
+# Ignore all PDFs (keep data exceptions above) and do not track demo PDFs
+# If you need to commit a specific demo PDF, remove this negation locally.
+# The following line used to force-add a large demo PDF; remove it to satisfy pre-commit:
+# !apps/multimodal/vision-based-pdf-multi-vector/pdfs/2004.12832v2.pdf
+!apps/multimodal/vision-based-pdf-multi-vector/fig/*

.gitmodules

@@ -14,3 +14,6 @@
 [submodule "packages/leann-backend-hnsw/third_party/libzmq"]
 	path = packages/leann-backend-hnsw/third_party/libzmq
 	url = https://github.com/zeromq/libzmq.git
+[submodule "packages/astchunk-leann"]
+	path = packages/astchunk-leann
+	url = https://github.com/yichuan-w/astchunk-leann.git

README.md

@@ -8,6 +8,8 @@
   <img src="https://img.shields.io/badge/Platform-Ubuntu%20%26%20Arch%20%26%20WSL%20%7C%20macOS%20(ARM64%2FIntel)-lightgrey" alt="Platform">
   <img src="https://img.shields.io/badge/License-MIT-green.svg" alt="MIT License">
   <img src="https://img.shields.io/badge/MCP-Native%20Integration-blue" alt="MCP Integration">
+  <a href="https://join.slack.com/t/leann-e2u9779/shared_invite/zt-3ckd2f6w1-OX08~NN4gkWhh10PRVBj1Q"><img src="https://img.shields.io/badge/Slack-Join-4A154B?logo=slack&logoColor=white" alt="Join Slack">
+  <a href="assets/wechat_user_group.JPG" title="Join WeChat group"><img src="https://img.shields.io/badge/WeChat-Join-2DC100?logo=wechat&logoColor=white" alt="Join WeChat group"></a>
 </p>
 
 <h2 align="center" tabindex="-1" class="heading-element" dir="auto">
@@ -18,7 +20,7 @@ LEANN is an innovative vector database that democratizes personal AI. Transform
 
 LEANN achieves this through *graph-based selective recomputation* with *high-degree preserving pruning*, computing embeddings on-demand instead of storing them all. [Illustration Fig →](#️-architecture--how-it-works) | [Paper →](https://arxiv.org/abs/2506.08276)
 
-**Ready to RAG Everything?** Transform your laptop into a personal AI assistant that can semantic search your **[file system](#-personal-data-manager-process-any-documents-pdf-txt-md)**, **[emails](#-your-personal-email-secretary-rag-on-apple-mail)**, **[browser history](#-time-machine-for-the-web-rag-your-entire-browser-history)**, **[chat history](#-wechat-detective-unlock-your-golden-memories)**, **[codebase](#-claude-code-integration-transform-your-development-workflow)**\* , or external knowledge bases (i.e., 60M documents) - all on your laptop, with zero cloud costs and complete privacy.
+**Ready to RAG Everything?** Transform your laptop into a personal AI assistant that can semantic search your **[file system](#-personal-data-manager-process-any-documents-pdf-txt-md)**, **[emails](#-your-personal-email-secretary-rag-on-apple-mail)**, **[browser history](#-time-machine-for-the-web-rag-your-entire-browser-history)**, **[chat history](#-wechat-detective-unlock-your-golden-memories)** ([WeChat](#-wechat-detective-unlock-your-golden-memories), [iMessage](#-imessage-history-your-personal-conversation-archive)), **[agent memory](#-chatgpt-chat-history-your-personal-ai-conversation-archive)** ([ChatGPT](#-chatgpt-chat-history-your-personal-ai-conversation-archive), [Claude](#-claude-chat-history-your-personal-ai-conversation-archive)), **[codebase](#-claude-code-integration-transform-your-development-workflow)**\* , or external knowledge bases (i.e., 60M documents) - all on your laptop, with zero cloud costs and complete privacy.
 
 
 \* Claude Code only supports basic `grep`-style keyword search. **LEANN** is a drop-in **semantic search MCP service fully compatible with Claude Code**, unlocking intelligent retrieval without changing your workflow. 🔥 Check out [the easy setup →](packages/leann-mcp/README.md)
@@ -174,13 +176,16 @@ response = chat.ask("How much storage does LEANN save?", top_k=1)
 
 ## RAG on Everything!
 
-LEANN supports RAG on various data sources including documents (`.pdf`, `.txt`, `.md`), Apple Mail, Google Search History, WeChat, and more.
+LEANN supports RAG on various data sources including documents (`.pdf`, `.txt`, `.md`), Apple Mail, Google Search History, WeChat, ChatGPT conversations, Claude conversations, iMessage conversations, and more.
 
 
 
 ### Generation Model Setup
 
-LEANN supports multiple LLM providers for text generation (OpenAI API, HuggingFace, Ollama).
+#### LLM Backend
+
+LEANN supports many LLM providers for text generation (HuggingFace, Ollama, and Any OpenAI compatible API).
+
 
 <details>
 <summary><strong>🔑 OpenAI API Setup (Default)</strong></summary>
@@ -191,6 +196,68 @@ Set your OpenAI API key as an environment variable:
 export OPENAI_API_KEY="your-api-key-here"
 ```
 
+Make sure to use `--llm openai` flag when using the CLI.
+You can also specify the model name with `--llm-model <model-name>` flag.
+
+</details>
+
+<details>
+<summary><strong>🛠️ Supported LLM & Embedding Providers (via OpenAI Compatibility)</strong></summary>
+
+Thanks to the widespread adoption of the OpenAI API format, LEANN is compatible out-of-the-box with a vast array of LLM and embedding providers. Simply set the `OPENAI_BASE_URL` and `OPENAI_API_KEY` environment variables to connect to your preferred service.
+
+```sh
+export OPENAI_API_KEY="xxx"
+export OPENAI_BASE_URL="http://localhost:1234/v1" # base url of the provider
+```
+
+To use OpenAI compatible endpoint with the CLI interface:
+
+If you are using it for text generation, make sure to use `--llm openai` flag and specify the model name with `--llm-model <model-name>` flag.
+
+If you are using it for embedding, set the `--embedding-mode openai` flag and specify the model name with `--embedding-model <MODEL>`.
+
+-----
+
+
+Below is a list of base URLs for common providers to get you started.
+
+
+### 🖥️ Local Inference Engines (Recommended for full privacy)
+
+| Provider         | Sample Base URL             |
+| ---------------- | --------------------------- |
+| **Ollama** | `http://localhost:11434/v1` |
+| **LM Studio** | `http://localhost:1234/v1`  |
+| **vLLM** | `http://localhost:8000/v1`  |
+| **llama.cpp** | `http://localhost:8080/v1`  |
+| **SGLang** | `http://localhost:30000/v1` |
+| **LiteLLM** | `http://localhost:4000`     |
+
+-----
+
+### ☁️ Cloud Providers
+
+> **🚨 A Note on Privacy:** Before choosing a cloud provider, carefully review their privacy and data retention policies. Depending on their terms, your data may be used for their own purposes, including but not limited to human reviews and model training, which can lead to serious consequences if not handled properly.
+
+
+| Provider         | Base URL                                                   |
+| ---------------- | ---------------------------------------------------------- |
+| **OpenAI** | `https://api.openai.com/v1`                                |
+| **OpenRouter** | `https://openrouter.ai/api/v1`                             |
+| **Gemini** | `https://generativelanguage.googleapis.com/v1beta/openai/` |
+| **x.AI (Grok)** | `https://api.x.ai/v1`                                      |
+| **Groq AI** | `https://api.groq.com/openai/v1`                           |
+| **DeepSeek** | `https://api.deepseek.com/v1`                              |
+| **SiliconFlow** | `https://api.siliconflow.cn/v1`                            |
+| **Zhipu (BigModel)** | `https://open.bigmodel.cn/api/paas/v4/`                |
+| **Mistral AI** | `https://api.mistral.ai/v1`                                |
+
+
+
+
+If your provider isn't on this list, don't worry! Check their documentation for an OpenAI-compatible endpoint—chances are, it's OpenAI Compatible too!
+
 </details>
 
 <details>
@@ -475,6 +542,238 @@ Once the index is built, you can ask questions like:
 
 </details>
 
+### 🤖 ChatGPT Chat History: Your Personal AI Conversation Archive!
+
+Transform your ChatGPT conversations into a searchable knowledge base! Search through all your ChatGPT discussions about coding, research, brainstorming, and more.
+
+```bash
+python -m apps.chatgpt_rag --export-path chatgpt_export.html --query "How do I create a list in Python?"
+```
+
+**Unlock your AI conversation history.** Never lose track of valuable insights from your ChatGPT discussions again.
+
+<details>
+<summary><strong>📋 Click to expand: How to Export ChatGPT Data</strong></summary>
+
+**Step-by-step export process:**
+
+1. **Sign in to ChatGPT**
+2. **Click your profile icon** in the top right corner
+3. **Navigate to Settings** → **Data Controls**
+4. **Click "Export"** under Export Data
+5. **Confirm the export** request
+6. **Download the ZIP file** from the email link (expires in 24 hours)
+7. **Extract or use directly** with LEANN
+
+**Supported formats:**
+- `.html` files from ChatGPT exports
+- `.zip` archives from ChatGPT
+- Directories with multiple export files
+
+</details>
+
+<details>
+<summary><strong>📋 Click to expand: ChatGPT-Specific Arguments</strong></summary>
+
+#### Parameters
+```bash
+--export-path PATH           # Path to ChatGPT export file (.html/.zip) or directory (default: ./chatgpt_export)
+--separate-messages         # Process each message separately instead of concatenated conversations
+--chunk-size N              # Text chunk size (default: 512)
+--chunk-overlap N           # Overlap between chunks (default: 128)
+```
+
+#### Example Commands
+```bash
+# Basic usage with HTML export
+python -m apps.chatgpt_rag --export-path conversations.html
+
+# Process ZIP archive from ChatGPT
+python -m apps.chatgpt_rag --export-path chatgpt_export.zip
+
+# Search with specific query
+python -m apps.chatgpt_rag --export-path chatgpt_data.html --query "Python programming help"
+
+# Process individual messages for fine-grained search
+python -m apps.chatgpt_rag --separate-messages --export-path chatgpt_export.html
+
+# Process directory containing multiple exports
+python -m apps.chatgpt_rag --export-path ./chatgpt_exports/ --max-items 1000
+```
+
+</details>
+
+<details>
+<summary><strong>💡 Click to expand: Example queries you can try</strong></summary>
+
+Once your ChatGPT conversations are indexed, you can search with queries like:
+- "What did I ask ChatGPT about Python programming?"
+- "Show me conversations about machine learning algorithms"
+- "Find discussions about web development frameworks"
+- "What coding advice did ChatGPT give me?"
+- "Search for conversations about debugging techniques"
+- "Find ChatGPT's recommendations for learning resources"
+
+</details>
+
+### 🤖 Claude Chat History: Your Personal AI Conversation Archive!
+
+Transform your Claude conversations into a searchable knowledge base! Search through all your Claude discussions about coding, research, brainstorming, and more.
+
+```bash
+python -m apps.claude_rag --export-path claude_export.json --query "What did I ask about Python dictionaries?"
+```
+
+**Unlock your AI conversation history.** Never lose track of valuable insights from your Claude discussions again.
+
+<details>
+<summary><strong>📋 Click to expand: How to Export Claude Data</strong></summary>
+
+**Step-by-step export process:**
+
+1. **Open Claude** in your browser
+2. **Navigate to Settings** (look for gear icon or settings menu)
+3. **Find Export/Download** options in your account settings
+4. **Download conversation data** (usually in JSON format)
+5. **Place the file** in your project directory
+
+*Note: Claude export methods may vary depending on the interface you're using. Check Claude's help documentation for the most current export instructions.*
+
+**Supported formats:**
+- `.json` files (recommended)
+- `.zip` archives containing JSON data
+- Directories with multiple export files
+
+</details>
+
+<details>
+<summary><strong>📋 Click to expand: Claude-Specific Arguments</strong></summary>
+
+#### Parameters
+```bash
+--export-path PATH           # Path to Claude export file (.json/.zip) or directory (default: ./claude_export)
+--separate-messages         # Process each message separately instead of concatenated conversations
+--chunk-size N              # Text chunk size (default: 512)
+--chunk-overlap N           # Overlap between chunks (default: 128)
+```
+
+#### Example Commands
+```bash
+# Basic usage with JSON export
+python -m apps.claude_rag --export-path my_claude_conversations.json
+
+# Process ZIP archive from Claude
+python -m apps.claude_rag --export-path claude_export.zip
+
+# Search with specific query
+python -m apps.claude_rag --export-path claude_data.json --query "machine learning advice"
+
+# Process individual messages for fine-grained search
+python -m apps.claude_rag --separate-messages --export-path claude_export.json
+
+# Process directory containing multiple exports
+python -m apps.claude_rag --export-path ./claude_exports/ --max-items 1000
+```
+
+</details>
+
+<details>
+<summary><strong>💡 Click to expand: Example queries you can try</strong></summary>
+
+Once your Claude conversations are indexed, you can search with queries like:
+- "What did I ask Claude about Python programming?"
+- "Show me conversations about machine learning algorithms"
+- "Find discussions about software architecture patterns"
+- "What debugging advice did Claude give me?"
+- "Search for conversations about data structures"
+- "Find Claude's recommendations for learning resources"
+
+</details>
+
+### 💬 iMessage History: Your Personal Conversation Archive!
+
+Transform your iMessage conversations into a searchable knowledge base! Search through all your text messages, group chats, and conversations with friends, family, and colleagues.
+
+```bash
+python -m apps.imessage_rag --query "What did we discuss about the weekend plans?"
+```
+
+**Unlock your message history.** Never lose track of important conversations, shared links, or memorable moments from your iMessage history.
+
+<details>
+<summary><strong>📋 Click to expand: How to Access iMessage Data</strong></summary>
+
+**iMessage data location:**
+
+iMessage conversations are stored in a SQLite database on your Mac at:
+```
+~/Library/Messages/chat.db
+```
+
+**Important setup requirements:**
+
+1. **Grant Full Disk Access** to your terminal or IDE:
+   - Open **System Preferences** → **Security & Privacy** → **Privacy**
+   - Select **Full Disk Access** from the left sidebar
+   - Click the **+** button and add your terminal app (Terminal, iTerm2) or IDE (VS Code, etc.)
+   - Restart your terminal/IDE after granting access
+
+2. **Alternative: Use a backup database**
+   - If you have Time Machine backups or manual copies of the database
+   - Use `--db-path` to specify a custom location
+
+**Supported formats:**
+- Direct access to `~/Library/Messages/chat.db` (default)
+- Custom database path with `--db-path`
+- Works with backup copies of the database
+
+</details>
+
+<details>
+<summary><strong>📋 Click to expand: iMessage-Specific Arguments</strong></summary>
+
+#### Parameters
+```bash
+--db-path PATH                    # Path to chat.db file (default: ~/Library/Messages/chat.db)
+--concatenate-conversations       # Group messages by conversation (default: True)
+--no-concatenate-conversations    # Process each message individually
+--chunk-size N                    # Text chunk size (default: 1000)
+--chunk-overlap N                 # Overlap between chunks (default: 200)
+```
+
+#### Example Commands
+```bash
+# Basic usage (requires Full Disk Access)
+python -m apps.imessage_rag
+
+# Search with specific query
+python -m apps.imessage_rag --query "family dinner plans"
+
+# Use custom database path
+python -m apps.imessage_rag --db-path /path/to/backup/chat.db
+
+# Process individual messages instead of conversations
+python -m apps.imessage_rag --no-concatenate-conversations
+
+# Limit processing for testing
+python -m apps.imessage_rag --max-items 100 --query "weekend"
+```
+
+</details>
+
+<details>
+<summary><strong>💡 Click to expand: Example queries you can try</strong></summary>
+
+Once your iMessage conversations are indexed, you can search with queries like:
+- "What did we discuss about vacation plans?"
+- "Find messages about restaurant recommendations"
+- "Show me conversations with John about the project"
+- "Search for shared links about technology"
+- "Find group chat discussions about weekend events"
+- "What did mom say about the family gathering?"
+
+</details>
+
 ### 🚀 Claude Code Integration: Transform Your Development Workflow!
 
 <details>
@@ -544,6 +843,9 @@ leann search my-docs "machine learning concepts"
 # Interactive chat with your documents
 leann ask my-docs --interactive
 
+# Ask a single question (non-interactive)
+leann ask my-docs "Where are prompts configured?"
+
 # List all your indexes
 leann list
 
@@ -654,6 +956,19 @@ results = searcher.search(
 
 📖 **[Complete Metadata filtering guide →](docs/metadata_filtering.md)**
 
+### 🔍 Grep Search
+
+For exact text matching instead of semantic search, use the `use_grep` parameter:
+
+```python
+# Exact text search
+results = searcher.search("banana‑crocodile", use_grep=True, top_k=1)
+```
+
+**Use cases**: Finding specific code patterns, error messages, function names, or exact phrases where semantic similarity isn't needed.
+
+📖 **[Complete grep search guide →](docs/grep_search.md)**
+
 ## 🏗️ Architecture & How It Works
 
 <p align="center">
@@ -691,9 +1006,8 @@ results = searcher.search(
 ## Reproduce Our Results
 
 ```bash
-uv pip install -e ".[dev]"  # Install dev dependencies
-python benchmarks/run_evaluation.py    # Will auto-download evaluation data and run benchmarks
-python benchmarks/run_evaluation.py benchmarks/data/indices/rpj_wiki/rpj_wiki --num-queries 2000    # After downloading data, you can run the benchmark with our biggest index
+uv run benchmarks/run_evaluation.py    # Will auto-download evaluation data and run benchmarks
+uv run benchmarks/run_evaluation.py benchmarks/data/indices/rpj_wiki/rpj_wiki --num-queries 2000    # After downloading data, you can run the benchmark with our biggest index
 ```
 
 The evaluation script downloads data automatically on first run. The last three results were tested with partial personal data, and you can reproduce them with your own data!

apps/base_rag_example.py

@@ -10,7 +10,9 @@ from typing import Any
 
 import dotenv
 from leann.api import LeannBuilder, LeannChat
+from leann.interactive_utils import create_rag_session
 from leann.registry import register_project_directory
+from leann.settings import resolve_ollama_host, resolve_openai_api_key, resolve_openai_base_url
 
 dotenv.load_dotenv()
 
@@ -78,6 +80,24 @@ class BaseRAGExample(ABC):
             choices=["sentence-transformers", "openai", "mlx", "ollama"],
             help="Embedding backend mode (default: sentence-transformers), we provide sentence-transformers, openai, mlx, or ollama",
         )
+        embedding_group.add_argument(
+            "--embedding-host",
+            type=str,
+            default=None,
+            help="Override Ollama-compatible embedding host",
+        )
+        embedding_group.add_argument(
+            "--embedding-api-base",
+            type=str,
+            default=None,
+            help="Base URL for OpenAI-compatible embedding services",
+        )
+        embedding_group.add_argument(
+            "--embedding-api-key",
+            type=str,
+            default=None,
+            help="API key for embedding service (defaults to OPENAI_API_KEY)",
+        )
 
         # LLM parameters
         llm_group = parser.add_argument_group("LLM Parameters")
@@ -97,8 +117,8 @@ class BaseRAGExample(ABC):
         llm_group.add_argument(
             "--llm-host",
             type=str,
-            default="http://localhost:11434",
-            help="Host for Ollama API (default: http://localhost:11434)",
+            default=None,
+            help="Host for Ollama-compatible APIs (defaults to LEANN_OLLAMA_HOST/OLLAMA_HOST)",
         )
         llm_group.add_argument(
             "--thinking-budget",
@@ -107,6 +127,18 @@ class BaseRAGExample(ABC):
             default=None,
             help="Thinking budget for reasoning models (low/medium/high). Supported by GPT-Oss:20b and other reasoning models.",
         )
+        llm_group.add_argument(
+            "--llm-api-base",
+            type=str,
+            default=None,
+            help="Base URL for OpenAI-compatible APIs",
+        )
+        llm_group.add_argument(
+            "--llm-api-key",
+            type=str,
+            default=None,
+            help="API key for OpenAI-compatible APIs (defaults to OPENAI_API_KEY)",
+        )
 
         # AST Chunking parameters
         ast_group = parser.add_argument_group("AST Chunking Parameters")
@@ -205,9 +237,13 @@ class BaseRAGExample(ABC):
 
         if args.llm == "openai":
             config["model"] = args.llm_model or "gpt-4o"
+            config["base_url"] = resolve_openai_base_url(args.llm_api_base)
+            resolved_key = resolve_openai_api_key(args.llm_api_key)
+            if resolved_key:
+                config["api_key"] = resolved_key
         elif args.llm == "ollama":
             config["model"] = args.llm_model or "llama3.2:1b"
-            config["host"] = args.llm_host
+            config["host"] = resolve_ollama_host(args.llm_host)
         elif args.llm == "hf":
             config["model"] = args.llm_model or "Qwen/Qwen2.5-1.5B-Instruct"
         elif args.llm == "simulated":
@@ -223,10 +259,20 @@ class BaseRAGExample(ABC):
         print(f"\n[Building Index] Creating {self.name} index...")
         print(f"Total text chunks: {len(texts)}")
 
+        embedding_options: dict[str, Any] = {}
+        if args.embedding_mode == "ollama":
+            embedding_options["host"] = resolve_ollama_host(args.embedding_host)
+        elif args.embedding_mode == "openai":
+            embedding_options["base_url"] = resolve_openai_base_url(args.embedding_api_base)
+            resolved_embedding_key = resolve_openai_api_key(args.embedding_api_key)
+            if resolved_embedding_key:
+                embedding_options["api_key"] = resolved_embedding_key
+
         builder = LeannBuilder(
             backend_name=args.backend_name,
             embedding_model=args.embedding_model,
             embedding_mode=args.embedding_mode,
+            embedding_options=embedding_options or None,
             graph_degree=args.graph_degree,
             complexity=args.build_complexity,
             is_compact=not args.no_compact,
@@ -262,19 +308,12 @@ class BaseRAGExample(ABC):
             complexity=args.search_complexity,
         )
 
-        print(f"\n[Interactive Mode] Chat with your {self.name} data!")
-        print("Type 'quit' or 'exit' to stop.\n")
-
-        while True:
-            try:
-                query = input("You: ").strip()
-                if query.lower() in ["quit", "exit", "q"]:
-                    print("Goodbye!")
-                    break
-
-                if not query:
-                    continue
+        # Create interactive session
+        session = create_rag_session(
+            app_name=self.name.lower().replace(" ", "_"), data_description=self.name
+        )
 
+        def handle_query(query: str):
             # Prepare LLM kwargs with thinking budget if specified
             llm_kwargs = {}
             if hasattr(args, "thinking_budget") and args.thinking_budget:
@@ -288,11 +327,7 @@ class BaseRAGExample(ABC):
             )
             print(f"\nAssistant: {response}\n")
 
-            except KeyboardInterrupt:
-                print("\nGoodbye!")
-                break
-            except Exception as e:
-                print(f"Error: {e}")
+        session.run_interactive_loop(handle_query)
 
     async def run_single_query(self, args, index_path: str, query: str):
         """Run a single query against the index."""

apps/browser_rag.py

@@ -10,7 +10,8 @@ from pathlib import Path
 # Add parent directory to path for imports
 sys.path.insert(0, str(Path(__file__).parent))
 
-from base_rag_example import BaseRAGExample, create_text_chunks
+from base_rag_example import BaseRAGExample
+from chunking import create_text_chunks
 
 from .history_data.history import ChromeHistoryReader
 

apps/chatgpt_data/chatgpt_reader.py

@@ -0,0 +1,413 @@
+"""
+ChatGPT export data reader.
+
+Reads and processes ChatGPT export data from chat.html files.
+"""
+
+import re
+from pathlib import Path
+from typing import Any
+from zipfile import ZipFile
+
+from bs4 import BeautifulSoup
+from llama_index.core import Document
+from llama_index.core.readers.base import BaseReader
+
+
+class ChatGPTReader(BaseReader):
+    """
+    ChatGPT export data reader.
+
+    Reads ChatGPT conversation data from exported chat.html files or zip archives.
+    Processes conversations into structured documents with metadata.
+    """
+
+    def __init__(self, concatenate_conversations: bool = True) -> None:
+        """
+        Initialize.
+
+        Args:
+            concatenate_conversations: Whether to concatenate messages within conversations for better context
+        """
+        try:
+            from bs4 import BeautifulSoup  # noqa
+        except ImportError:
+            raise ImportError("`beautifulsoup4` package not found: `pip install beautifulsoup4`")
+
+        self.concatenate_conversations = concatenate_conversations
+
+    def _extract_html_from_zip(self, zip_path: Path) -> str | None:
+        """
+        Extract chat.html from ChatGPT export zip file.
+
+        Args:
+            zip_path: Path to the ChatGPT export zip file
+
+        Returns:
+            HTML content as string, or None if not found
+        """
+        try:
+            with ZipFile(zip_path, "r") as zip_file:
+                # Look for chat.html or conversations.html
+                html_files = [
+                    f
+                    for f in zip_file.namelist()
+                    if f.endswith(".html") and ("chat" in f.lower() or "conversation" in f.lower())
+                ]
+
+                if not html_files:
+                    print(f"No HTML chat file found in {zip_path}")
+                    return None
+
+                # Use the first HTML file found
+                html_file = html_files[0]
+                print(f"Found HTML file: {html_file}")
+
+                with zip_file.open(html_file) as f:
+                    return f.read().decode("utf-8", errors="ignore")
+
+        except Exception as e:
+            print(f"Error extracting HTML from zip {zip_path}: {e}")
+            return None
+
+    def _parse_chatgpt_html(self, html_content: str) -> list[dict]:
+        """
+        Parse ChatGPT HTML export to extract conversations.
+
+        Args:
+            html_content: HTML content from ChatGPT export
+
+        Returns:
+            List of conversation dictionaries
+        """
+        soup = BeautifulSoup(html_content, "html.parser")
+        conversations = []
+
+        # Try different possible structures for ChatGPT exports
+        # Structure 1: Look for conversation containers
+        conversation_containers = soup.find_all(
+            ["div", "section"], class_=re.compile(r"conversation|chat", re.I)
+        )
+
+        if not conversation_containers:
+            # Structure 2: Look for message containers directly
+            conversation_containers = [soup]  # Use the entire document as one conversation
+
+        for container in conversation_containers:
+            conversation = self._extract_conversation_from_container(container)
+            if conversation and conversation.get("messages"):
+                conversations.append(conversation)
+
+        # If no structured conversations found, try to extract all text as one conversation
+        if not conversations:
+            all_text = soup.get_text(separator="\n", strip=True)
+            if all_text:
+                conversations.append(
+                    {
+                        "title": "ChatGPT Conversation",
+                        "messages": [{"role": "mixed", "content": all_text, "timestamp": None}],
+                        "timestamp": None,
+                    }
+                )
+
+        return conversations
+
+    def _extract_conversation_from_container(self, container) -> dict | None:
+        """
+        Extract conversation data from a container element.
+
+        Args:
+            container: BeautifulSoup element containing conversation
+
+        Returns:
+            Dictionary with conversation data or None
+        """
+        messages = []
+
+        # Look for message elements with various possible structures
+        message_selectors = ['[class*="message"]', '[class*="chat"]', "[data-message]", "p", "div"]
+
+        for selector in message_selectors:
+            message_elements = container.select(selector)
+            if message_elements:
+                break
+        else:
+            message_elements = []
+
+        # If no structured messages found, treat the entire container as one message
+        if not message_elements:
+            text_content = container.get_text(separator="\n", strip=True)
+            if text_content:
+                messages.append({"role": "mixed", "content": text_content, "timestamp": None})
+        else:
+            for element in message_elements:
+                message = self._extract_message_from_element(element)
+                if message:
+                    messages.append(message)
+
+        if not messages:
+            return None
+
+        # Try to extract conversation title
+        title_element = container.find(["h1", "h2", "h3", "title"])
+        title = title_element.get_text(strip=True) if title_element else "ChatGPT Conversation"
+
+        # Try to extract timestamp from various possible locations
+        timestamp = self._extract_timestamp_from_container(container)
+
+        return {"title": title, "messages": messages, "timestamp": timestamp}
+
+    def _extract_message_from_element(self, element) -> dict | None:
+        """
+        Extract message data from an element.
+
+        Args:
+            element: BeautifulSoup element containing message
+
+        Returns:
+            Dictionary with message data or None
+        """
+        text_content = element.get_text(separator=" ", strip=True)
+
+        # Skip empty or very short messages
+        if not text_content or len(text_content.strip()) < 3:
+            return None
+
+        # Try to determine role (user/assistant) from class names or content
+        role = "mixed"  # Default role
+
+        class_names = " ".join(element.get("class", [])).lower()
+        if "user" in class_names or "human" in class_names:
+            role = "user"
+        elif "assistant" in class_names or "ai" in class_names or "gpt" in class_names:
+            role = "assistant"
+        elif text_content.lower().startswith(("you:", "user:", "me:")):
+            role = "user"
+            text_content = re.sub(r"^(you|user|me):\s*", "", text_content, flags=re.IGNORECASE)
+        elif text_content.lower().startswith(("chatgpt:", "assistant:", "ai:")):
+            role = "assistant"
+            text_content = re.sub(
+                r"^(chatgpt|assistant|ai):\s*", "", text_content, flags=re.IGNORECASE
+            )
+
+        # Try to extract timestamp
+        timestamp = self._extract_timestamp_from_element(element)
+
+        return {"role": role, "content": text_content, "timestamp": timestamp}
+
+    def _extract_timestamp_from_element(self, element) -> str | None:
+        """Extract timestamp from element."""
+        # Look for timestamp in various attributes and child elements
+        timestamp_attrs = ["data-timestamp", "timestamp", "datetime"]
+        for attr in timestamp_attrs:
+            if element.get(attr):
+                return element.get(attr)
+
+        # Look for time elements
+        time_element = element.find("time")
+        if time_element:
+            return time_element.get("datetime") or time_element.get_text(strip=True)
+
+        # Look for date-like text patterns
+        text = element.get_text()
+        date_patterns = [r"\d{4}-\d{2}-\d{2}", r"\d{1,2}/\d{1,2}/\d{4}", r"\w+ \d{1,2}, \d{4}"]
+
+        for pattern in date_patterns:
+            match = re.search(pattern, text)
+            if match:
+                return match.group()
+
+        return None
+
+    def _extract_timestamp_from_container(self, container) -> str | None:
+        """Extract timestamp from conversation container."""
+        return self._extract_timestamp_from_element(container)
+
+    def _create_concatenated_content(self, conversation: dict) -> str:
+        """
+        Create concatenated content from conversation messages.
+
+        Args:
+            conversation: Dictionary containing conversation data
+
+        Returns:
+            Formatted concatenated content
+        """
+        title = conversation.get("title", "ChatGPT Conversation")
+        messages = conversation.get("messages", [])
+        timestamp = conversation.get("timestamp", "Unknown")
+
+        # Build message content
+        message_parts = []
+        for message in messages:
+            role = message.get("role", "mixed")
+            content = message.get("content", "")
+            msg_timestamp = message.get("timestamp", "")
+
+            if role == "user":
+                prefix = "[You]"
+            elif role == "assistant":
+                prefix = "[ChatGPT]"
+            else:
+                prefix = "[Message]"
+
+            # Add timestamp if available
+            if msg_timestamp:
+                prefix += f" ({msg_timestamp})"
+
+            message_parts.append(f"{prefix}: {content}")
+
+        concatenated_text = "\n\n".join(message_parts)
+
+        # Create final document content
+        doc_content = f"""Conversation: {title}
+Date: {timestamp}
+Messages ({len(messages)} messages):
+
+{concatenated_text}
+"""
+        return doc_content
+
+    def load_data(self, input_dir: str | None = None, **load_kwargs: Any) -> list[Document]:
+        """
+        Load ChatGPT export data.
+
+        Args:
+            input_dir: Directory containing ChatGPT export files or path to specific file
+            **load_kwargs:
+                max_count (int): Maximum number of conversations to process
+                chatgpt_export_path (str): Specific path to ChatGPT export file/directory
+                include_metadata (bool): Whether to include metadata in documents
+        """
+        docs: list[Document] = []
+        max_count = load_kwargs.get("max_count", -1)
+        chatgpt_export_path = load_kwargs.get("chatgpt_export_path", input_dir)
+        include_metadata = load_kwargs.get("include_metadata", True)
+
+        if not chatgpt_export_path:
+            print("No ChatGPT export path provided")
+            return docs
+
+        export_path = Path(chatgpt_export_path)
+
+        if not export_path.exists():
+            print(f"ChatGPT export path not found: {export_path}")
+            return docs
+
+        html_content = None
+
+        # Handle different input types
+        if export_path.is_file():
+            if export_path.suffix.lower() == ".zip":
+                # Extract HTML from zip file
+                html_content = self._extract_html_from_zip(export_path)
+            elif export_path.suffix.lower() == ".html":
+                # Read HTML file directly
+                try:
+                    with open(export_path, encoding="utf-8", errors="ignore") as f:
+                        html_content = f.read()
+                except Exception as e:
+                    print(f"Error reading HTML file {export_path}: {e}")
+                    return docs
+            else:
+                print(f"Unsupported file type: {export_path.suffix}")
+                return docs
+
+        elif export_path.is_dir():
+            # Look for HTML files in directory
+            html_files = list(export_path.glob("*.html"))
+            zip_files = list(export_path.glob("*.zip"))
+
+            if html_files:
+                # Use first HTML file found
+                html_file = html_files[0]
+                print(f"Found HTML file: {html_file}")
+                try:
+                    with open(html_file, encoding="utf-8", errors="ignore") as f:
+                        html_content = f.read()
+                except Exception as e:
+                    print(f"Error reading HTML file {html_file}: {e}")
+                    return docs
+
+            elif zip_files:
+                # Use first zip file found
+                zip_file = zip_files[0]
+                print(f"Found zip file: {zip_file}")
+                html_content = self._extract_html_from_zip(zip_file)
+
+            else:
+                print(f"No HTML or zip files found in {export_path}")
+                return docs
+
+        if not html_content:
+            print("No HTML content found to process")
+            return docs
+
+        # Parse conversations from HTML
+        print("Parsing ChatGPT conversations from HTML...")
+        conversations = self._parse_chatgpt_html(html_content)
+
+        if not conversations:
+            print("No conversations found in HTML content")
+            return docs
+
+        print(f"Found {len(conversations)} conversations")
+
+        # Process conversations into documents
+        count = 0
+        for conversation in conversations:
+            if max_count > 0 and count >= max_count:
+                break
+
+            if self.concatenate_conversations:
+                # Create one document per conversation with concatenated messages
+                doc_content = self._create_concatenated_content(conversation)
+
+                metadata = {}
+                if include_metadata:
+                    metadata = {
+                        "title": conversation.get("title", "ChatGPT Conversation"),
+                        "timestamp": conversation.get("timestamp", "Unknown"),
+                        "message_count": len(conversation.get("messages", [])),
+                        "source": "ChatGPT Export",
+                    }
+
+                doc = Document(text=doc_content, metadata=metadata)
+                docs.append(doc)
+                count += 1
+
+            else:
+                # Create separate documents for each message
+                for message in conversation.get("messages", []):
+                    if max_count > 0 and count >= max_count:
+                        break
+
+                    role = message.get("role", "mixed")
+                    content = message.get("content", "")
+                    msg_timestamp = message.get("timestamp", "")
+
+                    if not content.strip():
+                        continue
+
+                    # Create document content with context
+                    doc_content = f"""Conversation: {conversation.get("title", "ChatGPT Conversation")}
+Role: {role}
+Timestamp: {msg_timestamp or conversation.get("timestamp", "Unknown")}
+Message: {content}
+"""
+
+                    metadata = {}
+                    if include_metadata:
+                        metadata = {
+                            "conversation_title": conversation.get("title", "ChatGPT Conversation"),
+                            "role": role,
+                            "timestamp": msg_timestamp or conversation.get("timestamp", "Unknown"),
+                            "source": "ChatGPT Export",
+                        }
+
+                    doc = Document(text=doc_content, metadata=metadata)
+                    docs.append(doc)
+                    count += 1
+
+        print(f"Created {len(docs)} documents from ChatGPT export")
+        return docs

apps/chatgpt_rag.py

@@ -0,0 +1,186 @@
+"""
+ChatGPT RAG example using the unified interface.
+Supports ChatGPT export data from chat.html files.
+"""
+
+import sys
+from pathlib import Path
+
+# Add parent directory to path for imports
+sys.path.insert(0, str(Path(__file__).parent))
+
+from base_rag_example import BaseRAGExample
+from chunking import create_text_chunks
+
+from .chatgpt_data.chatgpt_reader import ChatGPTReader
+
+
+class ChatGPTRAG(BaseRAGExample):
+    """RAG example for ChatGPT conversation data."""
+
+    def __init__(self):
+        # Set default values BEFORE calling super().__init__
+        self.max_items_default = -1  # Process all conversations by default
+        self.embedding_model_default = (
+            "sentence-transformers/all-MiniLM-L6-v2"  # Fast 384-dim model
+        )
+
+        super().__init__(
+            name="ChatGPT",
+            description="Process and query ChatGPT conversation exports with LEANN",
+            default_index_name="chatgpt_conversations_index",
+        )
+
+    def _add_specific_arguments(self, parser):
+        """Add ChatGPT-specific arguments."""
+        chatgpt_group = parser.add_argument_group("ChatGPT Parameters")
+        chatgpt_group.add_argument(
+            "--export-path",
+            type=str,
+            default="./chatgpt_export",
+            help="Path to ChatGPT export file (.zip or .html) or directory containing exports (default: ./chatgpt_export)",
+        )
+        chatgpt_group.add_argument(
+            "--concatenate-conversations",
+            action="store_true",
+            default=True,
+            help="Concatenate messages within conversations for better context (default: True)",
+        )
+        chatgpt_group.add_argument(
+            "--separate-messages",
+            action="store_true",
+            help="Process each message as a separate document (overrides --concatenate-conversations)",
+        )
+        chatgpt_group.add_argument(
+            "--chunk-size", type=int, default=512, help="Text chunk size (default: 512)"
+        )
+        chatgpt_group.add_argument(
+            "--chunk-overlap", type=int, default=128, help="Text chunk overlap (default: 128)"
+        )
+
+    def _find_chatgpt_exports(self, export_path: Path) -> list[Path]:
+        """
+        Find ChatGPT export files in the given path.
+
+        Args:
+            export_path: Path to search for exports
+
+        Returns:
+            List of paths to ChatGPT export files
+        """
+        export_files = []
+
+        if export_path.is_file():
+            if export_path.suffix.lower() in [".zip", ".html"]:
+                export_files.append(export_path)
+        elif export_path.is_dir():
+            # Look for zip and html files
+            export_files.extend(export_path.glob("*.zip"))
+            export_files.extend(export_path.glob("*.html"))
+
+        return export_files
+
+    async def load_data(self, args) -> list[str]:
+        """Load ChatGPT export data and convert to text chunks."""
+        export_path = Path(args.export_path)
+
+        if not export_path.exists():
+            print(f"ChatGPT export path not found: {export_path}")
+            print(
+                "Please ensure you have exported your ChatGPT data and placed it in the correct location."
+            )
+            print("\nTo export your ChatGPT data:")
+            print("1. Sign in to ChatGPT")
+            print("2. Click on your profile icon → Settings → Data Controls")
+            print("3. Click 'Export' under Export Data")
+            print("4. Download the zip file from the email link")
+            print("5. Extract or place the file/directory at the specified path")
+            return []
+
+        # Find export files
+        export_files = self._find_chatgpt_exports(export_path)
+
+        if not export_files:
+            print(f"No ChatGPT export files (.zip or .html) found in: {export_path}")
+            return []
+
+        print(f"Found {len(export_files)} ChatGPT export files")
+
+        # Create reader with appropriate settings
+        concatenate = args.concatenate_conversations and not args.separate_messages
+        reader = ChatGPTReader(concatenate_conversations=concatenate)
+
+        # Process each export file
+        all_documents = []
+        total_processed = 0
+
+        for i, export_file in enumerate(export_files):
+            print(f"\nProcessing export file {i + 1}/{len(export_files)}: {export_file.name}")
+
+            try:
+                # Apply max_items limit per file
+                max_per_file = -1
+                if args.max_items > 0:
+                    remaining = args.max_items - total_processed
+                    if remaining <= 0:
+                        break
+                    max_per_file = remaining
+
+                # Load conversations
+                documents = reader.load_data(
+                    chatgpt_export_path=str(export_file),
+                    max_count=max_per_file,
+                    include_metadata=True,
+                )
+
+                if documents:
+                    all_documents.extend(documents)
+                    total_processed += len(documents)
+                    print(f"Processed {len(documents)} conversations from this file")
+                else:
+                    print(f"No conversations loaded from {export_file}")
+
+            except Exception as e:
+                print(f"Error processing {export_file}: {e}")
+                continue
+
+        if not all_documents:
+            print("No conversations found to process!")
+            print("\nTroubleshooting:")
+            print("- Ensure the export file is a valid ChatGPT export")
+            print("- Check that the HTML file contains conversation data")
+            print("- Try extracting the zip file and pointing to the HTML file directly")
+            return []
+
+        print(f"\nTotal conversations processed: {len(all_documents)}")
+        print("Now starting to split into text chunks... this may take some time")
+
+        # Convert to text chunks
+        all_texts = create_text_chunks(
+            all_documents, chunk_size=args.chunk_size, chunk_overlap=args.chunk_overlap
+        )
+
+        print(f"Created {len(all_texts)} text chunks from {len(all_documents)} conversations")
+        return all_texts
+
+
+if __name__ == "__main__":
+    import asyncio
+
+    # Example queries for ChatGPT RAG
+    print("\n🤖 ChatGPT RAG Example")
+    print("=" * 50)
+    print("\nExample queries you can try:")
+    print("- 'What did I ask about Python programming?'")
+    print("- 'Show me conversations about machine learning'")
+    print("- 'Find discussions about travel planning'")
+    print("- 'What advice did ChatGPT give me about career development?'")
+    print("- 'Search for conversations about cooking recipes'")
+    print("\nTo get started:")
+    print("1. Export your ChatGPT data from Settings → Data Controls → Export")
+    print("2. Place the downloaded zip file or extracted HTML in ./chatgpt_export/")
+    print("3. Run this script to build your personal ChatGPT knowledge base!")
+    print("\nOr run without --query for interactive mode\n")
+
+    rag = ChatGPTRAG()
+    asyncio.run(rag.run())

apps/chunking/__init__.py

@@ -1,9 +1,16 @@
-"""
-Chunking utilities for LEANN RAG applications.
-Provides AST-aware and traditional text chunking functionality.
+"""Unified chunking utilities facade.
+
+This module re-exports the packaged utilities from `leann.chunking_utils` so
+that both repo apps (importing `chunking`) and installed wheels share one
+single implementation. When running from the repo without installation, it
+adds the `packages/leann-core/src` directory to `sys.path` as a fallback.
 """
 
-from .utils import (
+import sys
+from pathlib import Path
+
+try:
+    from leann.chunking_utils import (
         CODE_EXTENSIONS,
         create_ast_chunks,
         create_text_chunks,
@@ -11,6 +18,21 @@ from .utils import (
         detect_code_files,
         get_language_from_extension,
     )
+except Exception:  # pragma: no cover - best-effort fallback for dev environment
+    repo_root = Path(__file__).resolve().parents[2]
+    leann_src = repo_root / "packages" / "leann-core" / "src"
+    if leann_src.exists():
+        sys.path.insert(0, str(leann_src))
+        from leann.chunking_utils import (
+            CODE_EXTENSIONS,
+            create_ast_chunks,
+            create_text_chunks,
+            create_traditional_chunks,
+            detect_code_files,
+            get_language_from_extension,
+        )
+    else:
+        raise
 
 __all__ = [
     "CODE_EXTENSIONS",

apps/claude_data/claude_reader.py

@@ -0,0 +1,420 @@
+"""
+Claude export data reader.
+
+Reads and processes Claude conversation data from exported JSON files.
+"""
+
+import json
+from pathlib import Path
+from typing import Any
+from zipfile import ZipFile
+
+from llama_index.core import Document
+from llama_index.core.readers.base import BaseReader
+
+
+class ClaudeReader(BaseReader):
+    """
+    Claude export data reader.
+
+    Reads Claude conversation data from exported JSON files or zip archives.
+    Processes conversations into structured documents with metadata.
+    """
+
+    def __init__(self, concatenate_conversations: bool = True) -> None:
+        """
+        Initialize.
+
+        Args:
+            concatenate_conversations: Whether to concatenate messages within conversations for better context
+        """
+        self.concatenate_conversations = concatenate_conversations
+
+    def _extract_json_from_zip(self, zip_path: Path) -> list[str]:
+        """
+        Extract JSON files from Claude export zip file.
+
+        Args:
+            zip_path: Path to the Claude export zip file
+
+        Returns:
+            List of JSON content strings, or empty list if not found
+        """
+        json_contents = []
+        try:
+            with ZipFile(zip_path, "r") as zip_file:
+                # Look for JSON files
+                json_files = [f for f in zip_file.namelist() if f.endswith(".json")]
+
+                if not json_files:
+                    print(f"No JSON files found in {zip_path}")
+                    return []
+
+                print(f"Found {len(json_files)} JSON files in archive")
+
+                for json_file in json_files:
+                    with zip_file.open(json_file) as f:
+                        content = f.read().decode("utf-8", errors="ignore")
+                        json_contents.append(content)
+
+        except Exception as e:
+            print(f"Error extracting JSON from zip {zip_path}: {e}")
+
+        return json_contents
+
+    def _parse_claude_json(self, json_content: str) -> list[dict]:
+        """
+        Parse Claude JSON export to extract conversations.
+
+        Args:
+            json_content: JSON content from Claude export
+
+        Returns:
+            List of conversation dictionaries
+        """
+        try:
+            data = json.loads(json_content)
+        except json.JSONDecodeError as e:
+            print(f"Error parsing JSON: {e}")
+            return []
+
+        conversations = []
+
+        # Handle different possible JSON structures
+        if isinstance(data, list):
+            # If data is a list of conversations
+            for item in data:
+                conversation = self._extract_conversation_from_json(item)
+                if conversation:
+                    conversations.append(conversation)
+        elif isinstance(data, dict):
+            # Check for common structures
+            if "conversations" in data:
+                # Structure: {"conversations": [...]}
+                for item in data["conversations"]:
+                    conversation = self._extract_conversation_from_json(item)
+                    if conversation:
+                        conversations.append(conversation)
+            elif "messages" in data:
+                # Single conversation with messages
+                conversation = self._extract_conversation_from_json(data)
+                if conversation:
+                    conversations.append(conversation)
+            else:
+                # Try to treat the whole object as a conversation
+                conversation = self._extract_conversation_from_json(data)
+                if conversation:
+                    conversations.append(conversation)
+
+        return conversations
+
+    def _extract_conversation_from_json(self, conv_data: dict) -> dict | None:
+        """
+        Extract conversation data from a JSON object.
+
+        Args:
+            conv_data: Dictionary containing conversation data
+
+        Returns:
+            Dictionary with conversation data or None
+        """
+        if not isinstance(conv_data, dict):
+            return None
+
+        messages = []
+
+        # Look for messages in various possible structures
+        message_sources = []
+        if "messages" in conv_data:
+            message_sources = conv_data["messages"]
+        elif "chat" in conv_data:
+            message_sources = conv_data["chat"]
+        elif "conversation" in conv_data:
+            message_sources = conv_data["conversation"]
+        else:
+            # If no clear message structure, try to extract from the object itself
+            if "content" in conv_data and "role" in conv_data:
+                message_sources = [conv_data]
+
+        for msg_data in message_sources:
+            message = self._extract_message_from_json(msg_data)
+            if message:
+                messages.append(message)
+
+        if not messages:
+            return None
+
+        # Extract conversation metadata
+        title = self._extract_title_from_conversation(conv_data, messages)
+        timestamp = self._extract_timestamp_from_conversation(conv_data)
+
+        return {"title": title, "messages": messages, "timestamp": timestamp}
+
+    def _extract_message_from_json(self, msg_data: dict) -> dict | None:
+        """
+        Extract message data from a JSON message object.
+
+        Args:
+            msg_data: Dictionary containing message data
+
+        Returns:
+            Dictionary with message data or None
+        """
+        if not isinstance(msg_data, dict):
+            return None
+
+        # Extract content from various possible fields
+        content = ""
+        content_fields = ["content", "text", "message", "body"]
+        for field in content_fields:
+            if msg_data.get(field):
+                content = str(msg_data[field])
+                break
+
+        if not content or len(content.strip()) < 3:
+            return None
+
+        # Extract role (user/assistant/human/ai/claude)
+        role = "mixed"  # Default role
+        role_fields = ["role", "sender", "from", "author", "type"]
+        for field in role_fields:
+            if msg_data.get(field):
+                role_value = str(msg_data[field]).lower()
+                if role_value in ["user", "human", "person"]:
+                    role = "user"
+                elif role_value in ["assistant", "ai", "claude", "bot"]:
+                    role = "assistant"
+                break
+
+        # Extract timestamp
+        timestamp = self._extract_timestamp_from_message(msg_data)
+
+        return {"role": role, "content": content, "timestamp": timestamp}
+
+    def _extract_timestamp_from_message(self, msg_data: dict) -> str | None:
+        """Extract timestamp from message data."""
+        timestamp_fields = ["timestamp", "created_at", "date", "time"]
+        for field in timestamp_fields:
+            if msg_data.get(field):
+                return str(msg_data[field])
+        return None
+
+    def _extract_timestamp_from_conversation(self, conv_data: dict) -> str | None:
+        """Extract timestamp from conversation data."""
+        timestamp_fields = ["timestamp", "created_at", "date", "updated_at", "last_updated"]
+        for field in timestamp_fields:
+            if conv_data.get(field):
+                return str(conv_data[field])
+        return None
+
+    def _extract_title_from_conversation(self, conv_data: dict, messages: list) -> str:
+        """Extract or generate title for conversation."""
+        # Try to find explicit title
+        title_fields = ["title", "name", "subject", "topic"]
+        for field in title_fields:
+            if conv_data.get(field):
+                return str(conv_data[field])
+
+        # Generate title from first user message
+        for message in messages:
+            if message.get("role") == "user":
+                content = message.get("content", "")
+                if content:
+                    # Use first 50 characters as title
+                    title = content[:50].strip()
+                    if len(content) > 50:
+                        title += "..."
+                    return title
+
+        return "Claude Conversation"
+
+    def _create_concatenated_content(self, conversation: dict) -> str:
+        """
+        Create concatenated content from conversation messages.
+
+        Args:
+            conversation: Dictionary containing conversation data
+
+        Returns:
+            Formatted concatenated content
+        """
+        title = conversation.get("title", "Claude Conversation")
+        messages = conversation.get("messages", [])
+        timestamp = conversation.get("timestamp", "Unknown")
+
+        # Build message content
+        message_parts = []
+        for message in messages:
+            role = message.get("role", "mixed")
+            content = message.get("content", "")
+            msg_timestamp = message.get("timestamp", "")
+
+            if role == "user":
+                prefix = "[You]"
+            elif role == "assistant":
+                prefix = "[Claude]"
+            else:
+                prefix = "[Message]"
+
+            # Add timestamp if available
+            if msg_timestamp:
+                prefix += f" ({msg_timestamp})"
+
+            message_parts.append(f"{prefix}: {content}")
+
+        concatenated_text = "\n\n".join(message_parts)
+
+        # Create final document content
+        doc_content = f"""Conversation: {title}
+Date: {timestamp}
+Messages ({len(messages)} messages):
+
+{concatenated_text}
+"""
+        return doc_content
+
+    def load_data(self, input_dir: str | None = None, **load_kwargs: Any) -> list[Document]:
+        """
+        Load Claude export data.
+
+        Args:
+            input_dir: Directory containing Claude export files or path to specific file
+            **load_kwargs:
+                max_count (int): Maximum number of conversations to process
+                claude_export_path (str): Specific path to Claude export file/directory
+                include_metadata (bool): Whether to include metadata in documents
+        """
+        docs: list[Document] = []
+        max_count = load_kwargs.get("max_count", -1)
+        claude_export_path = load_kwargs.get("claude_export_path", input_dir)
+        include_metadata = load_kwargs.get("include_metadata", True)
+
+        if not claude_export_path:
+            print("No Claude export path provided")
+            return docs
+
+        export_path = Path(claude_export_path)
+
+        if not export_path.exists():
+            print(f"Claude export path not found: {export_path}")
+            return docs
+
+        json_contents = []
+
+        # Handle different input types
+        if export_path.is_file():
+            if export_path.suffix.lower() == ".zip":
+                # Extract JSON from zip file
+                json_contents = self._extract_json_from_zip(export_path)
+            elif export_path.suffix.lower() == ".json":
+                # Read JSON file directly
+                try:
+                    with open(export_path, encoding="utf-8", errors="ignore") as f:
+                        json_contents.append(f.read())
+                except Exception as e:
+                    print(f"Error reading JSON file {export_path}: {e}")
+                    return docs
+            else:
+                print(f"Unsupported file type: {export_path.suffix}")
+                return docs
+
+        elif export_path.is_dir():
+            # Look for JSON files in directory
+            json_files = list(export_path.glob("*.json"))
+            zip_files = list(export_path.glob("*.zip"))
+
+            if json_files:
+                print(f"Found {len(json_files)} JSON files in directory")
+                for json_file in json_files:
+                    try:
+                        with open(json_file, encoding="utf-8", errors="ignore") as f:
+                            json_contents.append(f.read())
+                    except Exception as e:
+                        print(f"Error reading JSON file {json_file}: {e}")
+                        continue
+
+            if zip_files:
+                print(f"Found {len(zip_files)} ZIP files in directory")
+                for zip_file in zip_files:
+                    zip_contents = self._extract_json_from_zip(zip_file)
+                    json_contents.extend(zip_contents)
+
+            if not json_files and not zip_files:
+                print(f"No JSON or ZIP files found in {export_path}")
+                return docs
+
+        if not json_contents:
+            print("No JSON content found to process")
+            return docs
+
+        # Parse conversations from JSON content
+        print("Parsing Claude conversations from JSON...")
+        all_conversations = []
+        for json_content in json_contents:
+            conversations = self._parse_claude_json(json_content)
+            all_conversations.extend(conversations)
+
+        if not all_conversations:
+            print("No conversations found in JSON content")
+            return docs
+
+        print(f"Found {len(all_conversations)} conversations")
+
+        # Process conversations into documents
+        count = 0
+        for conversation in all_conversations:
+            if max_count > 0 and count >= max_count:
+                break
+
+            if self.concatenate_conversations:
+                # Create one document per conversation with concatenated messages
+                doc_content = self._create_concatenated_content(conversation)
+
+                metadata = {}
+                if include_metadata:
+                    metadata = {
+                        "title": conversation.get("title", "Claude Conversation"),
+                        "timestamp": conversation.get("timestamp", "Unknown"),
+                        "message_count": len(conversation.get("messages", [])),
+                        "source": "Claude Export",
+                    }
+
+                doc = Document(text=doc_content, metadata=metadata)
+                docs.append(doc)
+                count += 1
+
+            else:
+                # Create separate documents for each message
+                for message in conversation.get("messages", []):
+                    if max_count > 0 and count >= max_count:
+                        break
+
+                    role = message.get("role", "mixed")
+                    content = message.get("content", "")
+                    msg_timestamp = message.get("timestamp", "")
+
+                    if not content.strip():
+                        continue
+
+                    # Create document content with context
+                    doc_content = f"""Conversation: {conversation.get("title", "Claude Conversation")}
+Role: {role}
+Timestamp: {msg_timestamp or conversation.get("timestamp", "Unknown")}
+Message: {content}
+"""
+
+                    metadata = {}
+                    if include_metadata:
+                        metadata = {
+                            "conversation_title": conversation.get("title", "Claude Conversation"),
+                            "role": role,
+                            "timestamp": msg_timestamp or conversation.get("timestamp", "Unknown"),
+                            "source": "Claude Export",
+                        }
+
+                    doc = Document(text=doc_content, metadata=metadata)
+                    docs.append(doc)
+                    count += 1
+
+        print(f"Created {len(docs)} documents from Claude export")
+        return docs

apps/claude_rag.py

@@ -0,0 +1,189 @@
+"""
+Claude RAG example using the unified interface.
+Supports Claude export data from JSON files.
+"""
+
+import sys
+from pathlib import Path
+
+# Add parent directory to path for imports
+sys.path.insert(0, str(Path(__file__).parent))
+
+from base_rag_example import BaseRAGExample
+from chunking import create_text_chunks
+
+from .claude_data.claude_reader import ClaudeReader
+
+
+class ClaudeRAG(BaseRAGExample):
+    """RAG example for Claude conversation data."""
+
+    def __init__(self):
+        # Set default values BEFORE calling super().__init__
+        self.max_items_default = -1  # Process all conversations by default
+        self.embedding_model_default = (
+            "sentence-transformers/all-MiniLM-L6-v2"  # Fast 384-dim model
+        )
+
+        super().__init__(
+            name="Claude",
+            description="Process and query Claude conversation exports with LEANN",
+            default_index_name="claude_conversations_index",
+        )
+
+    def _add_specific_arguments(self, parser):
+        """Add Claude-specific arguments."""
+        claude_group = parser.add_argument_group("Claude Parameters")
+        claude_group.add_argument(
+            "--export-path",
+            type=str,
+            default="./claude_export",
+            help="Path to Claude export file (.json or .zip) or directory containing exports (default: ./claude_export)",
+        )
+        claude_group.add_argument(
+            "--concatenate-conversations",
+            action="store_true",
+            default=True,
+            help="Concatenate messages within conversations for better context (default: True)",
+        )
+        claude_group.add_argument(
+            "--separate-messages",
+            action="store_true",
+            help="Process each message as a separate document (overrides --concatenate-conversations)",
+        )
+        claude_group.add_argument(
+            "--chunk-size", type=int, default=512, help="Text chunk size (default: 512)"
+        )
+        claude_group.add_argument(
+            "--chunk-overlap", type=int, default=128, help="Text chunk overlap (default: 128)"
+        )
+
+    def _find_claude_exports(self, export_path: Path) -> list[Path]:
+        """
+        Find Claude export files in the given path.
+
+        Args:
+            export_path: Path to search for exports
+
+        Returns:
+            List of paths to Claude export files
+        """
+        export_files = []
+
+        if export_path.is_file():
+            if export_path.suffix.lower() in [".zip", ".json"]:
+                export_files.append(export_path)
+        elif export_path.is_dir():
+            # Look for zip and json files
+            export_files.extend(export_path.glob("*.zip"))
+            export_files.extend(export_path.glob("*.json"))
+
+        return export_files
+
+    async def load_data(self, args) -> list[str]:
+        """Load Claude export data and convert to text chunks."""
+        export_path = Path(args.export_path)
+
+        if not export_path.exists():
+            print(f"Claude export path not found: {export_path}")
+            print(
+                "Please ensure you have exported your Claude data and placed it in the correct location."
+            )
+            print("\nTo export your Claude data:")
+            print("1. Open Claude in your browser")
+            print("2. Look for export/download options in settings or conversation menu")
+            print("3. Download the conversation data (usually in JSON format)")
+            print("4. Place the file/directory at the specified path")
+            print(
+                "\nNote: Claude export methods may vary. Check Claude's help documentation for current instructions."
+            )
+            return []
+
+        # Find export files
+        export_files = self._find_claude_exports(export_path)
+
+        if not export_files:
+            print(f"No Claude export files (.json or .zip) found in: {export_path}")
+            return []
+
+        print(f"Found {len(export_files)} Claude export files")
+
+        # Create reader with appropriate settings
+        concatenate = args.concatenate_conversations and not args.separate_messages
+        reader = ClaudeReader(concatenate_conversations=concatenate)
+
+        # Process each export file
+        all_documents = []
+        total_processed = 0
+
+        for i, export_file in enumerate(export_files):
+            print(f"\nProcessing export file {i + 1}/{len(export_files)}: {export_file.name}")
+
+            try:
+                # Apply max_items limit per file
+                max_per_file = -1
+                if args.max_items > 0:
+                    remaining = args.max_items - total_processed
+                    if remaining <= 0:
+                        break
+                    max_per_file = remaining
+
+                # Load conversations
+                documents = reader.load_data(
+                    claude_export_path=str(export_file),
+                    max_count=max_per_file,
+                    include_metadata=True,
+                )
+
+                if documents:
+                    all_documents.extend(documents)
+                    total_processed += len(documents)
+                    print(f"Processed {len(documents)} conversations from this file")
+                else:
+                    print(f"No conversations loaded from {export_file}")
+
+            except Exception as e:
+                print(f"Error processing {export_file}: {e}")
+                continue
+
+        if not all_documents:
+            print("No conversations found to process!")
+            print("\nTroubleshooting:")
+            print("- Ensure the export file is a valid Claude export")
+            print("- Check that the JSON file contains conversation data")
+            print("- Try using a different export format or method")
+            print("- Check Claude's documentation for current export procedures")
+            return []
+
+        print(f"\nTotal conversations processed: {len(all_documents)}")
+        print("Now starting to split into text chunks... this may take some time")
+
+        # Convert to text chunks
+        all_texts = create_text_chunks(
+            all_documents, chunk_size=args.chunk_size, chunk_overlap=args.chunk_overlap
+        )
+
+        print(f"Created {len(all_texts)} text chunks from {len(all_documents)} conversations")
+        return all_texts
+
+
+if __name__ == "__main__":
+    import asyncio
+
+    # Example queries for Claude RAG
+    print("\n🤖 Claude RAG Example")
+    print("=" * 50)
+    print("\nExample queries you can try:")
+    print("- 'What did I ask Claude about Python programming?'")
+    print("- 'Show me conversations about machine learning'")
+    print("- 'Find discussions about code optimization'")
+    print("- 'What advice did Claude give me about software design?'")
+    print("- 'Search for conversations about debugging techniques'")
+    print("\nTo get started:")
+    print("1. Export your Claude conversation data")
+    print("2. Place the JSON/ZIP file in ./claude_export/")
+    print("3. Run this script to build your personal Claude knowledge base!")
+    print("\nOr run without --query for interactive mode\n")
+
+    rag = ClaudeRAG()
+    asyncio.run(rag.run())

apps/email_rag.py

@@ -9,7 +9,8 @@ from pathlib import Path
 # Add parent directory to path for imports
 sys.path.insert(0, str(Path(__file__).parent))
 
-from base_rag_example import BaseRAGExample, create_text_chunks
+from base_rag_example import BaseRAGExample
+from chunking import create_text_chunks
 
 from .email_data.LEANN_email_reader import EmlxReader
 

apps/history_data/history.py

@@ -74,7 +74,7 @@ class ChromeHistoryReader(BaseReader):
                 if count >= max_count and max_count > 0:
                     break
 
-                last_visit, url, title, visit_count, typed_count, hidden = row
+                last_visit, url, title, visit_count, typed_count, _hidden = row
 
                 # Create document content with metadata embedded in text
                 doc_content = f"""

apps/imessage_data/__init__.py

@@ -0,0 +1 @@
+"""iMessage data processing module."""

apps/imessage_data/imessage_reader.py

@@ -0,0 +1,342 @@
+"""
+iMessage data reader.
+
+Reads and processes iMessage conversation data from the macOS Messages database.
+"""
+
+import sqlite3
+from datetime import datetime
+from pathlib import Path
+from typing import Any
+
+from llama_index.core import Document
+from llama_index.core.readers.base import BaseReader
+
+
+class IMessageReader(BaseReader):
+    """
+    iMessage data reader.
+
+    Reads iMessage conversation data from the macOS Messages database (chat.db).
+    Processes conversations into structured documents with metadata.
+    """
+
+    def __init__(self, concatenate_conversations: bool = True) -> None:
+        """
+        Initialize.
+
+        Args:
+            concatenate_conversations: Whether to concatenate messages within conversations for better context
+        """
+        self.concatenate_conversations = concatenate_conversations
+
+    def _get_default_chat_db_path(self) -> Path:
+        """
+        Get the default path to the iMessage chat database.
+
+        Returns:
+            Path to the chat.db file
+        """
+        home = Path.home()
+        return home / "Library" / "Messages" / "chat.db"
+
+    def _convert_cocoa_timestamp(self, cocoa_timestamp: int) -> str:
+        """
+        Convert Cocoa timestamp to readable format.
+
+        Args:
+            cocoa_timestamp: Timestamp in Cocoa format (nanoseconds since 2001-01-01)
+
+        Returns:
+            Formatted timestamp string
+        """
+        if cocoa_timestamp == 0:
+            return "Unknown"
+
+        try:
+            # Cocoa timestamp is nanoseconds since 2001-01-01 00:00:00 UTC
+            # Convert to seconds and add to Unix epoch
+            cocoa_epoch = datetime(2001, 1, 1)
+            unix_timestamp = cocoa_timestamp / 1_000_000_000  # Convert nanoseconds to seconds
+            message_time = cocoa_epoch.timestamp() + unix_timestamp
+            return datetime.fromtimestamp(message_time).strftime("%Y-%m-%d %H:%M:%S")
+        except (ValueError, OSError):
+            return "Unknown"
+
+    def _get_contact_name(self, handle_id: str) -> str:
+        """
+        Get a readable contact name from handle ID.
+
+        Args:
+            handle_id: The handle ID (phone number or email)
+
+        Returns:
+            Formatted contact name
+        """
+        if not handle_id:
+            return "Unknown"
+
+        # Clean up phone numbers and emails for display
+        if "@" in handle_id:
+            return handle_id  # Email address
+        elif handle_id.startswith("+"):
+            return handle_id  # International phone number
+        else:
+            # Try to format as phone number
+            digits = "".join(filter(str.isdigit, handle_id))
+            if len(digits) == 10:
+                return f"({digits[:3]}) {digits[3:6]}-{digits[6:]}"
+            elif len(digits) == 11 and digits[0] == "1":
+                return f"+1 ({digits[1:4]}) {digits[4:7]}-{digits[7:]}"
+            else:
+                return handle_id
+
+    def _read_messages_from_db(self, db_path: Path) -> list[dict]:
+        """
+        Read messages from the iMessage database.
+
+        Args:
+            db_path: Path to the chat.db file
+
+        Returns:
+            List of message dictionaries
+        """
+        if not db_path.exists():
+            print(f"iMessage database not found at: {db_path}")
+            return []
+
+        try:
+            # Connect to the database
+            conn = sqlite3.connect(str(db_path))
+            cursor = conn.cursor()
+
+            # Query to get messages with chat and handle information
+            query = """
+            SELECT
+                m.ROWID as message_id,
+                m.text,
+                m.date,
+                m.is_from_me,
+                m.service,
+                c.chat_identifier,
+                c.display_name as chat_display_name,
+                h.id as handle_id,
+                c.ROWID as chat_id
+            FROM message m
+            LEFT JOIN chat_message_join cmj ON m.ROWID = cmj.message_id
+            LEFT JOIN chat c ON cmj.chat_id = c.ROWID
+            LEFT JOIN handle h ON m.handle_id = h.ROWID
+            WHERE m.text IS NOT NULL AND m.text != ''
+            ORDER BY c.ROWID, m.date
+            """
+
+            cursor.execute(query)
+            rows = cursor.fetchall()
+
+            messages = []
+            for row in rows:
+                (
+                    message_id,
+                    text,
+                    date,
+                    is_from_me,
+                    service,
+                    chat_identifier,
+                    chat_display_name,
+                    handle_id,
+                    chat_id,
+                ) = row
+
+                message = {
+                    "message_id": message_id,
+                    "text": text,
+                    "timestamp": self._convert_cocoa_timestamp(date),
+                    "is_from_me": bool(is_from_me),
+                    "service": service or "iMessage",
+                    "chat_identifier": chat_identifier or "Unknown",
+                    "chat_display_name": chat_display_name or "Unknown Chat",
+                    "handle_id": handle_id or "Unknown",
+                    "contact_name": self._get_contact_name(handle_id or ""),
+                    "chat_id": chat_id,
+                }
+                messages.append(message)
+
+            conn.close()
+            print(f"Found {len(messages)} messages in database")
+            return messages
+
+        except sqlite3.Error as e:
+            print(f"Error reading iMessage database: {e}")
+            return []
+        except Exception as e:
+            print(f"Unexpected error reading iMessage database: {e}")
+            return []
+
+    def _group_messages_by_chat(self, messages: list[dict]) -> dict[int, list[dict]]:
+        """
+        Group messages by chat ID.
+
+        Args:
+            messages: List of message dictionaries
+
+        Returns:
+            Dictionary mapping chat_id to list of messages
+        """
+        chats = {}
+        for message in messages:
+            chat_id = message["chat_id"]
+            if chat_id not in chats:
+                chats[chat_id] = []
+            chats[chat_id].append(message)
+
+        return chats
+
+    def _create_concatenated_content(self, chat_id: int, messages: list[dict]) -> str:
+        """
+        Create concatenated content from chat messages.
+
+        Args:
+            chat_id: The chat ID
+            messages: List of messages in the chat
+
+        Returns:
+            Concatenated text content
+        """
+        if not messages:
+            return ""
+
+        # Get chat info from first message
+        first_msg = messages[0]
+        chat_name = first_msg["chat_display_name"]
+        chat_identifier = first_msg["chat_identifier"]
+
+        # Build message content
+        message_parts = []
+        for message in messages:
+            timestamp = message["timestamp"]
+            is_from_me = message["is_from_me"]
+            text = message["text"]
+            contact_name = message["contact_name"]
+
+            if is_from_me:
+                prefix = "[You]"
+            else:
+                prefix = f"[{contact_name}]"
+
+            if timestamp != "Unknown":
+                prefix += f" ({timestamp})"
+
+            message_parts.append(f"{prefix}: {text}")
+
+        concatenated_text = "\n\n".join(message_parts)
+
+        doc_content = f"""Chat: {chat_name}
+Identifier: {chat_identifier}
+Messages ({len(messages)} messages):
+
+{concatenated_text}
+"""
+        return doc_content
+
+    def _create_individual_content(self, message: dict) -> str:
+        """
+        Create content for individual message.
+
+        Args:
+            message: Message dictionary
+
+        Returns:
+            Formatted message content
+        """
+        timestamp = message["timestamp"]
+        is_from_me = message["is_from_me"]
+        text = message["text"]
+        contact_name = message["contact_name"]
+        chat_name = message["chat_display_name"]
+
+        sender = "You" if is_from_me else contact_name
+
+        return f"""Message from {sender} in chat "{chat_name}"
+Time: {timestamp}
+Content: {text}
+"""
+
+    def load_data(self, input_dir: str | None = None, **load_kwargs: Any) -> list[Document]:
+        """
+        Load iMessage data and return as documents.
+
+        Args:
+            input_dir: Optional path to directory containing chat.db file.
+                      If not provided, uses default macOS location.
+            **load_kwargs: Additional arguments (unused)
+
+        Returns:
+            List of Document objects containing iMessage data
+        """
+        docs = []
+
+        # Determine database path
+        if input_dir:
+            db_path = Path(input_dir) / "chat.db"
+        else:
+            db_path = self._get_default_chat_db_path()
+
+        print(f"Reading iMessage database from: {db_path}")
+
+        # Read messages from database
+        messages = self._read_messages_from_db(db_path)
+        if not messages:
+            return docs
+
+        if self.concatenate_conversations:
+            # Group messages by chat and create concatenated documents
+            chats = self._group_messages_by_chat(messages)
+
+            for chat_id, chat_messages in chats.items():
+                if not chat_messages:
+                    continue
+
+                content = self._create_concatenated_content(chat_id, chat_messages)
+
+                # Create metadata
+                first_msg = chat_messages[0]
+                last_msg = chat_messages[-1]
+
+                metadata = {
+                    "source": "iMessage",
+                    "chat_id": chat_id,
+                    "chat_name": first_msg["chat_display_name"],
+                    "chat_identifier": first_msg["chat_identifier"],
+                    "message_count": len(chat_messages),
+                    "first_message_date": first_msg["timestamp"],
+                    "last_message_date": last_msg["timestamp"],
+                    "participants": list(
+                        {msg["contact_name"] for msg in chat_messages if not msg["is_from_me"]}
+                    ),
+                }
+
+                doc = Document(text=content, metadata=metadata)
+                docs.append(doc)
+
+        else:
+            # Create individual documents for each message
+            for message in messages:
+                content = self._create_individual_content(message)
+
+                metadata = {
+                    "source": "iMessage",
+                    "message_id": message["message_id"],
+                    "chat_id": message["chat_id"],
+                    "chat_name": message["chat_display_name"],
+                    "chat_identifier": message["chat_identifier"],
+                    "timestamp": message["timestamp"],
+                    "is_from_me": message["is_from_me"],
+                    "contact_name": message["contact_name"],
+                    "service": message["service"],
+                }
+
+                doc = Document(text=content, metadata=metadata)
+                docs.append(doc)
+
+        print(f"Created {len(docs)} documents from iMessage data")
+        return docs

apps/imessage_rag.py

@@ -0,0 +1,125 @@
+"""
+iMessage RAG Example.
+
+This example demonstrates how to build a RAG system on your iMessage conversation history.
+"""
+
+import asyncio
+from pathlib import Path
+
+from leann.chunking_utils import create_text_chunks
+
+from apps.base_rag_example import BaseRAGExample
+from apps.imessage_data.imessage_reader import IMessageReader
+
+
+class IMessageRAG(BaseRAGExample):
+    """RAG example for iMessage conversation history."""
+
+    def __init__(self):
+        super().__init__(
+            name="iMessage",
+            description="RAG on your iMessage conversation history",
+            default_index_name="imessage_index",
+        )
+
+    def _add_specific_arguments(self, parser):
+        """Add iMessage-specific arguments."""
+        imessage_group = parser.add_argument_group("iMessage Parameters")
+        imessage_group.add_argument(
+            "--db-path",
+            type=str,
+            default=None,
+            help="Path to iMessage chat.db file (default: ~/Library/Messages/chat.db)",
+        )
+        imessage_group.add_argument(
+            "--concatenate-conversations",
+            action="store_true",
+            default=True,
+            help="Concatenate messages within conversations for better context (default: True)",
+        )
+        imessage_group.add_argument(
+            "--no-concatenate-conversations",
+            action="store_true",
+            help="Process each message individually instead of concatenating by conversation",
+        )
+        imessage_group.add_argument(
+            "--chunk-size",
+            type=int,
+            default=1000,
+            help="Maximum characters per text chunk (default: 1000)",
+        )
+        imessage_group.add_argument(
+            "--chunk-overlap",
+            type=int,
+            default=200,
+            help="Overlap between text chunks (default: 200)",
+        )
+
+    async def load_data(self, args) -> list[str]:
+        """Load iMessage history and convert to text chunks."""
+        print("Loading iMessage conversation history...")
+
+        # Determine concatenation setting
+        concatenate = args.concatenate_conversations and not args.no_concatenate_conversations
+
+        # Initialize iMessage reader
+        reader = IMessageReader(concatenate_conversations=concatenate)
+
+        # Load documents
+        try:
+            if args.db_path:
+                # Use custom database path
+                db_dir = str(Path(args.db_path).parent)
+                documents = reader.load_data(input_dir=db_dir)
+            else:
+                # Use default macOS location
+                documents = reader.load_data()
+
+        except Exception as e:
+            print(f"Error loading iMessage data: {e}")
+            print("\nTroubleshooting tips:")
+            print("1. Make sure you have granted Full Disk Access to your terminal/IDE")
+            print("2. Check that the iMessage database exists at ~/Library/Messages/chat.db")
+            print("3. Try specifying a custom path with --db-path if you have a backup")
+            return []
+
+        if not documents:
+            print("No iMessage conversations found!")
+            return []
+
+        print(f"Loaded {len(documents)} iMessage documents")
+
+        # Show some statistics
+        total_messages = sum(doc.metadata.get("message_count", 1) for doc in documents)
+        print(f"Total messages: {total_messages}")
+
+        if concatenate:
+            # Show chat statistics
+            chat_names = [doc.metadata.get("chat_name", "Unknown") for doc in documents]
+            unique_chats = len(set(chat_names))
+            print(f"Unique conversations: {unique_chats}")
+
+        # Convert to text chunks
+        all_texts = create_text_chunks(
+            documents,
+            chunk_size=args.chunk_size,
+            chunk_overlap=args.chunk_overlap,
+        )
+
+        # Apply max_items limit if specified
+        if args.max_items > 0:
+            all_texts = all_texts[: args.max_items]
+            print(f"Limited to {len(all_texts)} text chunks (max_items={args.max_items})")
+
+        return all_texts
+
+
+async def main():
+    """Main entry point."""
+    app = IMessageRAG()
+    await app.run()
+
+
+if __name__ == "__main__":
+    asyncio.run(main())

apps/multimodal/vision-based-pdf-multi-vector/README.md

@@ -0,0 +1,113 @@
+## Vision-based PDF Multi-Vector Demos (macOS/MPS)
+
+This folder contains two demos to index PDF pages as images and run multi-vector retrieval with ColPali/ColQwen2, plus optional similarity map visualization and answer generation.
+
+### What you’ll run
+- `multi-vector-leann-paper-example.py`: local PDF → pages → embed → build HNSW index → search.
+- `multi-vector-leann-similarity-map.py`: HF dataset (default) or local pages → embed → index → retrieve → similarity maps → optional Qwen-VL answer.
+
+## Prerequisites (macOS)
+
+### 1) Homebrew poppler (for pdf2image)
+```bash
+brew install poppler
+which pdfinfo && pdfinfo -v
+```
+
+### 2) Python environment
+Use uv (recommended) or pip. Python 3.9+.
+
+Using uv:
+```bash
+uv pip install \
+  colpali_engine \
+  pdf2image \
+  pillow \
+  matplotlib qwen_vl_utils \
+  einops \
+  seaborn
+```
+
+Notes:
+- On first run, models download from Hugging Face. Login/config if needed.
+- The scripts auto-select device: CUDA > MPS > CPU. Verify MPS:
+```bash
+python -c "import torch; print('MPS available:', bool(getattr(torch.backends, 'mps', None) and torch.backends.mps.is_available()))"
+```
+
+## Run the demos
+
+### A) Local PDF example
+Converts a local PDF into page images, embeds them, builds an index, and searches.
+
+```bash
+cd apps/multimodal/vision-based-pdf-multi-vector
+# If you don't have the sample PDF locally, download it (ignored by Git)
+mkdir -p pdfs
+curl -L -o pdfs/2004.12832v2.pdf https://arxiv.org/pdf/2004.12832.pdf
+ls pdfs/2004.12832v2.pdf
+# Ensure output dir exists
+mkdir -p pages
+python multi-vector-leann-paper-example.py
+```
+Expected:
+- Page images in `pages/`.
+- Console prints like `Using device=mps, dtype=...` and retrieved file paths for queries.
+
+To use your own PDF: edit `pdf_path` near the top of the script.
+
+### B) Similarity map + answer demo
+Uses HF dataset `weaviate/arXiv-AI-papers-multi-vector` by default; can switch to local pages.
+
+```bash
+cd apps/multimodal/vision-based-pdf-multi-vector
+python multi-vector-leann-similarity-map.py
+```
+Artifacts (when enabled):
+- Retrieved pages: `./figures/retrieved_page_rank{K}.png`
+- Similarity maps: `./figures/similarity_map_rank{K}.png`
+
+Key knobs in the script (top of file):
+- `QUERY`: your question
+- `MODEL`: `"colqwen2"` or `"colpali"`
+- `USE_HF_DATASET`: set `False` to use local pages
+- `PDF`, `PAGES_DIR`: for local mode
+- `INDEX_PATH`, `TOPK`, `FIRST_STAGE_K`, `REBUILD_INDEX`
+- `SIMILARITY_MAP`, `SIM_TOKEN_IDX`, `SIM_OUTPUT`
+- `ANSWER`, `MAX_NEW_TOKENS` (Qwen-VL)
+
+## Troubleshooting
+- pdf2image errors on macOS: ensure `brew install poppler` and `pdfinfo` works in terminal.
+- Slow or OOM on MPS: reduce dataset size (e.g., set `MAX_DOCS`) or switch to CPU.
+- NaNs on MPS: keep fp32 on MPS (default in similarity-map script); avoid fp16 there.
+- First-run model downloads can be large; ensure network access (HF mirrors if needed).
+
+## Notes
+- Index files are under `./indexes/`. Delete or set `REBUILD_INDEX=True` to rebuild.
+- For local PDFs, page images go to `./pages/`.
+
+
+### Retrieval and Visualization Example
+
+Example settings in `multi-vector-leann-similarity-map.py`:
+- `QUERY = "How does DeepSeek-V2 compare against the LLaMA family of LLMs?"`
+- `SIMILARITY_MAP = True` (to generate heatmaps)
+- `TOPK = 1` (save the top retrieved page and its similarity map)
+
+Run:
+```bash
+cd apps/multimodal/vision-based-pdf-multi-vector
+python multi-vector-leann-similarity-map.py
+```
+
+Outputs (by default):
+- Retrieved page: `./figures/retrieved_page_rank1.png`
+- Similarity map: `./figures/similarity_map_rank1.png`
+
+Sample visualization (example result, and the query is "QUERY = "How does Vim model performance and efficiency compared to other models?"
+"):
+![Similarity map example](fig/image.png)
+
+Notes:
+- Set `SIM_TOKEN_IDX` to visualize a specific token index; set `-1` to auto-select the most salient token.
+- If you change `SIM_OUTPUT` to a file path (e.g., `./figures/my_map.png`), multiple ranks are saved as `my_map_rank{K}.png`.

apps/multimodal/vision-based-pdf-multi-vector/leann_multi_vector.py

@@ -0,0 +1,182 @@
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+import numpy as np
+
+
+def _ensure_repo_paths_importable(current_file: str) -> None:
+    _repo_root = Path(current_file).resolve().parents[3]
+    _leann_core_src = _repo_root / "packages" / "leann-core" / "src"
+    _leann_hnsw_pkg = _repo_root / "packages" / "leann-backend-hnsw"
+    if str(_leann_core_src) not in sys.path:
+        sys.path.append(str(_leann_core_src))
+    if str(_leann_hnsw_pkg) not in sys.path:
+        sys.path.append(str(_leann_hnsw_pkg))
+
+
+_ensure_repo_paths_importable(__file__)
+
+from leann_backend_hnsw.hnsw_backend import HNSWBuilder, HNSWSearcher  # noqa: E402
+
+
+class LeannMultiVector:
+    def __init__(
+        self,
+        index_path: str,
+        dim: int = 128,
+        distance_metric: str = "mips",
+        m: int = 16,
+        ef_construction: int = 500,
+        is_compact: bool = False,
+        is_recompute: bool = False,
+        embedding_model_name: str = "colvision",
+    ) -> None:
+        self.index_path = index_path
+        self.dim = dim
+        self.embedding_model_name = embedding_model_name
+        self._pending_items: list[dict] = []
+        self._backend_kwargs = {
+            "distance_metric": distance_metric,
+            "M": m,
+            "efConstruction": ef_construction,
+            "is_compact": is_compact,
+            "is_recompute": is_recompute,
+        }
+        self._labels_meta: list[dict] = []
+
+    def _meta_dict(self) -> dict:
+        return {
+            "version": "1.0",
+            "backend_name": "hnsw",
+            "embedding_model": self.embedding_model_name,
+            "embedding_mode": "custom",
+            "dimensions": self.dim,
+            "backend_kwargs": self._backend_kwargs,
+            "is_compact": self._backend_kwargs.get("is_compact", True),
+            "is_pruned": self._backend_kwargs.get("is_compact", True)
+            and self._backend_kwargs.get("is_recompute", True),
+        }
+
+    def create_collection(self) -> None:
+        path = Path(self.index_path)
+        path.parent.mkdir(parents=True, exist_ok=True)
+
+    def insert(self, data: dict) -> None:
+        self._pending_items.append(
+            {
+                "doc_id": int(data["doc_id"]),
+                "filepath": data.get("filepath", ""),
+                "colbert_vecs": [np.asarray(v, dtype=np.float32) for v in data["colbert_vecs"]],
+            }
+        )
+
+    def _labels_path(self) -> Path:
+        index_path_obj = Path(self.index_path)
+        return index_path_obj.parent / f"{index_path_obj.name}.labels.json"
+
+    def _meta_path(self) -> Path:
+        index_path_obj = Path(self.index_path)
+        return index_path_obj.parent / f"{index_path_obj.name}.meta.json"
+
+    def create_index(self) -> None:
+        if not self._pending_items:
+            return
+
+        embeddings: list[np.ndarray] = []
+        labels_meta: list[dict] = []
+
+        for item in self._pending_items:
+            doc_id = int(item["doc_id"])
+            filepath = item.get("filepath", "")
+            colbert_vecs = item["colbert_vecs"]
+            for seq_id, vec in enumerate(colbert_vecs):
+                vec_np = np.asarray(vec, dtype=np.float32)
+                embeddings.append(vec_np)
+                labels_meta.append(
+                    {
+                        "id": f"{doc_id}:{seq_id}",
+                        "doc_id": doc_id,
+                        "seq_id": int(seq_id),
+                        "filepath": filepath,
+                    }
+                )
+
+        if not embeddings:
+            return
+
+        embeddings_np = np.vstack(embeddings).astype(np.float32)
+        # print shape of embeddings_np
+        print(embeddings_np.shape)
+
+        builder = HNSWBuilder(**{**self._backend_kwargs, "dimensions": self.dim})
+        ids = [str(i) for i in range(embeddings_np.shape[0])]
+        builder.build(embeddings_np, ids, self.index_path)
+
+        import json as _json
+
+        with open(self._meta_path(), "w", encoding="utf-8") as f:
+            _json.dump(self._meta_dict(), f, indent=2)
+        with open(self._labels_path(), "w", encoding="utf-8") as f:
+            _json.dump(labels_meta, f)
+
+        self._labels_meta = labels_meta
+
+    def _load_labels_meta_if_needed(self) -> None:
+        if self._labels_meta:
+            return
+        labels_path = self._labels_path()
+        if labels_path.exists():
+            import json as _json
+
+            with open(labels_path, encoding="utf-8") as f:
+                self._labels_meta = _json.load(f)
+
+    def search(
+        self, data: np.ndarray, topk: int, first_stage_k: int = 50
+    ) -> list[tuple[float, int]]:
+        if data.ndim == 1:
+            data = data.reshape(1, -1)
+        if data.dtype != np.float32:
+            data = data.astype(np.float32)
+
+        self._load_labels_meta_if_needed()
+
+        searcher = HNSWSearcher(self.index_path, meta=self._meta_dict())
+        raw = searcher.search(
+            data,
+            first_stage_k,
+            recompute_embeddings=False,
+            complexity=128,
+            beam_width=1,
+            prune_ratio=0.0,
+            batch_size=0,
+        )
+
+        labels = raw.get("labels")
+        distances = raw.get("distances")
+        if labels is None or distances is None:
+            return []
+
+        doc_scores: dict[int, float] = {}
+        B = len(labels)
+        for b in range(B):
+            per_doc_best: dict[int, float] = {}
+            for k, sid in enumerate(labels[b]):
+                try:
+                    idx = int(sid)
+                except Exception:
+                    continue
+                if 0 <= idx < len(self._labels_meta):
+                    doc_id = int(self._labels_meta[idx]["doc_id"])  # type: ignore[index]
+                else:
+                    continue
+                score = float(distances[b][k])
+                if (doc_id not in per_doc_best) or (score > per_doc_best[doc_id]):
+                    per_doc_best[doc_id] = score
+            for doc_id, best_score in per_doc_best.items():
+                doc_scores[doc_id] = doc_scores.get(doc_id, 0.0) + best_score
+
+        scores = sorted(((v, k) for k, v in doc_scores.items()), key=lambda x: x[0], reverse=True)
+        return scores[:topk] if len(scores) >= topk else scores

apps/multimodal/vision-based-pdf-multi-vector/multi-vector-leann-paper-example.py

@@ -0,0 +1,112 @@
+# pip install pdf2image
+# pip install pymilvus
+# pip install colpali_engine
+# pip install tqdm
+# pip install pillow
+
+import os
+import re
+import sys
+from pathlib import Path
+from typing import cast
+
+from PIL import Image
+from tqdm import tqdm
+
+# Ensure local leann packages are importable before importing them
+_repo_root = Path(__file__).resolve().parents[3]
+_leann_core_src = _repo_root / "packages" / "leann-core" / "src"
+_leann_hnsw_pkg = _repo_root / "packages" / "leann-backend-hnsw"
+if str(_leann_core_src) not in sys.path:
+    sys.path.append(str(_leann_core_src))
+if str(_leann_hnsw_pkg) not in sys.path:
+    sys.path.append(str(_leann_hnsw_pkg))
+
+
+import torch
+from colpali_engine.models import ColPali
+from colpali_engine.models.paligemma.colpali.processing_colpali import ColPaliProcessor
+from colpali_engine.utils.torch_utils import ListDataset, get_torch_device
+from torch.utils.data import DataLoader
+
+# Auto-select device: CUDA > MPS (mac) > CPU
+_device_str = (
+    "cuda"
+    if torch.cuda.is_available()
+    else (
+        "mps"
+        if getattr(torch.backends, "mps", None) and torch.backends.mps.is_available()
+        else "cpu"
+    )
+)
+device = get_torch_device(_device_str)
+# Prefer fp16 on GPU/MPS, bfloat16 on CPU
+_dtype = torch.float16 if _device_str in ("cuda", "mps") else torch.bfloat16
+model_name = "vidore/colpali-v1.2"
+
+model = ColPali.from_pretrained(
+    model_name,
+    torch_dtype=_dtype,
+    device_map=device,
+).eval()
+print(f"Using device={_device_str}, dtype={_dtype}")
+
+queries = [
+    "How to end-to-end retrieval with ColBert",
+    "Where is ColBERT performance Table, including text representation results?",
+]
+
+processor = cast(ColPaliProcessor, ColPaliProcessor.from_pretrained(model_name))
+
+dataloader = DataLoader(
+    dataset=ListDataset[str](queries),
+    batch_size=1,
+    shuffle=False,
+    collate_fn=lambda x: processor.process_queries(x),
+)
+
+qs: list[torch.Tensor] = []
+for batch_query in dataloader:
+    with torch.no_grad():
+        batch_query = {k: v.to(model.device) for k, v in batch_query.items()}
+        embeddings_query = model(**batch_query)
+    qs.extend(list(torch.unbind(embeddings_query.to("cpu"))))
+print(qs[0].shape)
+# %%
+page_filenames = sorted(os.listdir("./pages"), key=lambda n: int(re.search(r"\d+", n).group()))
+images = [Image.open(os.path.join("./pages", name)) for name in page_filenames]
+
+dataloader = DataLoader(
+    dataset=ListDataset[str](images),
+    batch_size=1,
+    shuffle=False,
+    collate_fn=lambda x: processor.process_images(x),
+)
+
+ds: list[torch.Tensor] = []
+for batch_doc in tqdm(dataloader):
+    with torch.no_grad():
+        batch_doc = {k: v.to(model.device) for k, v in batch_doc.items()}
+        embeddings_doc = model(**batch_doc)
+    ds.extend(list(torch.unbind(embeddings_doc.to("cpu"))))
+
+print(ds[0].shape)
+
+# %%
+# Build HNSW index via LeannRetriever primitives and run search
+index_path = "./indexes/colpali.leann"
+retriever = LeannRetriever(index_path=index_path, dim=int(ds[0].shape[-1]))
+retriever.create_collection()
+filepaths = [os.path.join("./pages", name) for name in page_filenames]
+for i in range(len(filepaths)):
+    data = {
+        "colbert_vecs": ds[i].float().numpy(),
+        "doc_id": i,
+        "filepath": filepaths[i],
+    }
+    retriever.insert(data)
+retriever.create_index()
+for query in qs:
+    query_np = query.float().numpy()
+    result = retriever.search(query_np, topk=1)
+    print(filepaths[result[0][1]])

apps/multimodal/vision-based-pdf-multi-vector/multi-vector-leann-similarity-map.py

@@ -0,0 +1,477 @@
+## Jupyter-style notebook script
+# %%
+# uv pip install matplotlib qwen_vl_utils
+import os
+import re
+import sys
+from pathlib import Path
+from typing import Any, Optional, cast
+
+from PIL import Image
+from tqdm import tqdm
+
+
+def _ensure_repo_paths_importable(current_file: str) -> None:
+    """Make local leann packages importable without installing (mirrors multi-vector-leann.py)."""
+    _repo_root = Path(current_file).resolve().parents[3]
+    _leann_core_src = _repo_root / "packages" / "leann-core" / "src"
+    _leann_hnsw_pkg = _repo_root / "packages" / "leann-backend-hnsw"
+    if str(_leann_core_src) not in sys.path:
+        sys.path.append(str(_leann_core_src))
+    if str(_leann_hnsw_pkg) not in sys.path:
+        sys.path.append(str(_leann_hnsw_pkg))
+
+
+_ensure_repo_paths_importable(__file__)
+
+from leann_multi_vector import LeannMultiVector  # noqa: E402
+
+# %%
+# Config
+os.environ["TOKENIZERS_PARALLELISM"] = "false"
+QUERY = "How does DeepSeek-V2 compare against the LLaMA family of LLMs?"
+MODEL: str = "colqwen2"  # "colpali" or "colqwen2"
+
+# Data source: set to True to use the Hugging Face dataset example (recommended)
+USE_HF_DATASET: bool = True
+DATASET_NAME: str = "weaviate/arXiv-AI-papers-multi-vector"
+DATASET_SPLIT: str = "train"
+MAX_DOCS: Optional[int] = None  # limit number of pages to index; None = all
+
+# Local pages (used when USE_HF_DATASET == False)
+PDF: Optional[str] = None  # e.g., "./pdfs/2004.12832v2.pdf"
+PAGES_DIR: str = "./pages"
+
+# Index + retrieval settings
+INDEX_PATH: str = "./indexes/colvision.leann"
+TOPK: int = 1
+FIRST_STAGE_K: int = 500
+REBUILD_INDEX: bool = False
+
+# Artifacts
+SAVE_TOP_IMAGE: Optional[str] = "./figures/retrieved_page.png"
+SIMILARITY_MAP: bool = True
+SIM_TOKEN_IDX: int = 13  # -1 means auto-select the most salient token
+SIM_OUTPUT: str = "./figures/similarity_map.png"
+ANSWER: bool = True
+MAX_NEW_TOKENS: int = 128
+
+
+# %%
+# Helpers
+def _natural_sort_key(name: str) -> int:
+    m = re.search(r"\d+", name)
+    return int(m.group()) if m else 0
+
+
+def _load_images_from_dir(pages_dir: str) -> tuple[list[str], list[Image.Image]]:
+    filenames = [n for n in os.listdir(pages_dir) if n.lower().endswith((".png", ".jpg", ".jpeg"))]
+    filenames = sorted(filenames, key=_natural_sort_key)
+    filepaths = [os.path.join(pages_dir, n) for n in filenames]
+    images = [Image.open(p) for p in filepaths]
+    return filepaths, images
+
+
+def _maybe_convert_pdf_to_images(pdf_path: Optional[str], pages_dir: str, dpi: int = 200) -> None:
+    if not pdf_path:
+        return
+    os.makedirs(pages_dir, exist_ok=True)
+    try:
+        from pdf2image import convert_from_path
+    except Exception as e:
+        raise RuntimeError(
+            "pdf2image is required to convert PDF to images. Install via pip install pdf2image"
+        ) from e
+    images = convert_from_path(pdf_path, dpi=dpi)
+    for i, image in enumerate(images):
+        image.save(os.path.join(pages_dir, f"page_{i + 1}.png"), "PNG")
+
+
+def _select_device_and_dtype():
+    import torch
+    from colpali_engine.utils.torch_utils import get_torch_device
+
+    device_str = (
+        "cuda"
+        if torch.cuda.is_available()
+        else (
+            "mps"
+            if getattr(torch.backends, "mps", None) and torch.backends.mps.is_available()
+            else "cpu"
+        )
+    )
+    device = get_torch_device(device_str)
+    # Stable dtype selection to avoid NaNs:
+    # - CUDA: prefer bfloat16 if supported, else float16
+    # - MPS: use float32 (fp16 on MPS can produce NaNs in some ops)
+    # - CPU: float32
+    if device_str == "cuda":
+        dtype = torch.bfloat16 if torch.cuda.is_bf16_supported() else torch.float16
+        try:
+            torch.backends.cuda.matmul.allow_tf32 = True  # Better stability/perf on Ampere+
+        except Exception:
+            pass
+    elif device_str == "mps":
+        dtype = torch.float32
+    else:
+        dtype = torch.float32
+    return device_str, device, dtype
+
+
+def _load_colvision(model_choice: str):
+    import torch
+    from colpali_engine.models import ColPali, ColQwen2, ColQwen2Processor
+    from colpali_engine.models.paligemma.colpali.processing_colpali import ColPaliProcessor
+    from transformers.utils.import_utils import is_flash_attn_2_available
+
+    device_str, device, dtype = _select_device_and_dtype()
+
+    if model_choice == "colqwen2":
+        model_name = "vidore/colqwen2-v1.0"
+        # On CPU/MPS we must avoid flash-attn and stay eager; on CUDA prefer flash-attn if available
+        attn_implementation = (
+            "flash_attention_2"
+            if (device_str == "cuda" and is_flash_attn_2_available())
+            else "eager"
+        )
+        model = ColQwen2.from_pretrained(
+            model_name,
+            torch_dtype=torch.bfloat16,
+            device_map=device,
+            attn_implementation=attn_implementation,
+        ).eval()
+        processor = ColQwen2Processor.from_pretrained(model_name)
+    else:
+        model_name = "vidore/colpali-v1.2"
+        model = ColPali.from_pretrained(
+            model_name,
+            torch_dtype=torch.bfloat16,
+            device_map=device,
+        ).eval()
+        processor = cast(ColPaliProcessor, ColPaliProcessor.from_pretrained(model_name))
+
+    return model_name, model, processor, device_str, device, dtype
+
+
+def _embed_images(model, processor, images: list[Image.Image]) -> list[Any]:
+    import torch
+    from colpali_engine.utils.torch_utils import ListDataset
+    from torch.utils.data import DataLoader
+
+    # Ensure deterministic eval and autocast for stability
+    model.eval()
+
+    dataloader = DataLoader(
+        dataset=ListDataset[Image.Image](images),
+        batch_size=1,
+        shuffle=False,
+        collate_fn=lambda x: processor.process_images(x),
+    )
+
+    doc_vecs: list[Any] = []
+    for batch_doc in tqdm(dataloader, desc="Embedding images"):
+        with torch.no_grad():
+            batch_doc = {k: v.to(model.device) for k, v in batch_doc.items()}
+            # autocast on CUDA for bf16/fp16; on CPU/MPS stay in fp32
+            if model.device.type == "cuda":
+                with torch.autocast(
+                    device_type="cuda",
+                    dtype=model.dtype if model.dtype.is_floating_point else torch.bfloat16,
+                ):
+                    embeddings_doc = model(**batch_doc)
+            else:
+                embeddings_doc = model(**batch_doc)
+        doc_vecs.extend(list(torch.unbind(embeddings_doc.to("cpu"))))
+    return doc_vecs
+
+
+def _embed_queries(model, processor, queries: list[str]) -> list[Any]:
+    import torch
+    from colpali_engine.utils.torch_utils import ListDataset
+    from torch.utils.data import DataLoader
+
+    model.eval()
+
+    dataloader = DataLoader(
+        dataset=ListDataset[str](queries),
+        batch_size=1,
+        shuffle=False,
+        collate_fn=lambda x: processor.process_queries(x),
+    )
+
+    q_vecs: list[Any] = []
+    for batch_query in tqdm(dataloader, desc="Embedding queries"):
+        with torch.no_grad():
+            batch_query = {k: v.to(model.device) for k, v in batch_query.items()}
+            if model.device.type == "cuda":
+                with torch.autocast(
+                    device_type="cuda",
+                    dtype=model.dtype if model.dtype.is_floating_point else torch.bfloat16,
+                ):
+                    embeddings_query = model(**batch_query)
+            else:
+                embeddings_query = model(**batch_query)
+        q_vecs.extend(list(torch.unbind(embeddings_query.to("cpu"))))
+    return q_vecs
+
+
+def _build_index(index_path: str, doc_vecs: list[Any], filepaths: list[str]) -> LeannMultiVector:
+    dim = int(doc_vecs[0].shape[-1])
+    retriever = LeannMultiVector(index_path=index_path, dim=dim)
+    retriever.create_collection()
+    for i, vec in enumerate(doc_vecs):
+        data = {
+            "colbert_vecs": vec.float().numpy(),
+            "doc_id": i,
+            "filepath": filepaths[i],
+        }
+        retriever.insert(data)
+    retriever.create_index()
+    return retriever
+
+
+def _load_retriever_if_index_exists(index_path: str, dim: int) -> Optional[LeannMultiVector]:
+    index_base = Path(index_path)
+    # Rough heuristic: index dir exists AND meta+labels files exist
+    meta = index_base.parent / f"{index_base.name}.meta.json"
+    labels = index_base.parent / f"{index_base.name}.labels.json"
+    if index_base.exists() and meta.exists() and labels.exists():
+        return LeannMultiVector(index_path=index_path, dim=dim)
+    return None
+
+
+def _generate_similarity_map(
+    model,
+    processor,
+    image: Image.Image,
+    query: str,
+    token_idx: Optional[int] = None,
+    output_path: Optional[str] = None,
+) -> tuple[int, float]:
+    import torch
+    from colpali_engine.interpretability import (
+        get_similarity_maps_from_embeddings,
+        plot_similarity_map,
+    )
+
+    batch_images = processor.process_images([image]).to(model.device)
+    batch_queries = processor.process_queries([query]).to(model.device)
+
+    with torch.no_grad():
+        image_embeddings = model.forward(**batch_images)
+        query_embeddings = model.forward(**batch_queries)
+
+    n_patches = processor.get_n_patches(
+        image_size=image.size,
+        spatial_merge_size=getattr(model, "spatial_merge_size", None),
+    )
+    image_mask = processor.get_image_mask(batch_images)
+
+    batched_similarity_maps = get_similarity_maps_from_embeddings(
+        image_embeddings=image_embeddings,
+        query_embeddings=query_embeddings,
+        n_patches=n_patches,
+        image_mask=image_mask,
+    )
+
+    similarity_maps = batched_similarity_maps[0]
+
+    # Determine token index if not provided: choose the token with highest max score
+    if token_idx is None:
+        per_token_max = similarity_maps.view(similarity_maps.shape[0], -1).max(dim=1).values
+        token_idx = int(per_token_max.argmax().item())
+
+    max_sim_score = similarity_maps[token_idx, :, :].max().item()
+
+    if output_path:
+        import matplotlib.pyplot as plt
+
+        fig, ax = plot_similarity_map(
+            image=image,
+            similarity_map=similarity_maps[token_idx],
+            figsize=(14, 14),
+            show_colorbar=False,
+        )
+        ax.set_title(f"Token #{token_idx}. MaxSim score: {max_sim_score:.2f}", fontsize=12)
+        os.makedirs(os.path.dirname(output_path), exist_ok=True)
+        plt.savefig(output_path, bbox_inches="tight")
+        plt.close(fig)
+
+    return token_idx, float(max_sim_score)
+
+
+class QwenVL:
+    def __init__(self, device: str):
+        from transformers import AutoProcessor, Qwen2_5_VLForConditionalGeneration
+        from transformers.utils.import_utils import is_flash_attn_2_available
+
+        attn_implementation = "flash_attention_2" if is_flash_attn_2_available() else "eager"
+        self.model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
+            "Qwen/Qwen2.5-VL-3B-Instruct",
+            torch_dtype="auto",
+            device_map=device,
+            attn_implementation=attn_implementation,
+        )
+
+        min_pixels = 256 * 28 * 28
+        max_pixels = 1280 * 28 * 28
+        self.processor = AutoProcessor.from_pretrained(
+            "Qwen/Qwen2.5-VL-3B-Instruct", min_pixels=min_pixels, max_pixels=max_pixels
+        )
+
+    def answer(self, query: str, images: list[Image.Image], max_new_tokens: int = 128) -> str:
+        import base64
+        from io import BytesIO
+
+        from qwen_vl_utils import process_vision_info
+
+        content = []
+        for img in images:
+            buffer = BytesIO()
+            img.save(buffer, format="jpeg")
+            img_base64 = base64.b64encode(buffer.getvalue()).decode("utf-8")
+            content.append({"type": "image", "image": f"data:image;base64,{img_base64}"})
+        content.append({"type": "text", "text": query})
+        messages = [{"role": "user", "content": content}]
+
+        text = self.processor.apply_chat_template(
+            messages, tokenize=False, add_generation_prompt=True
+        )
+        image_inputs, video_inputs = process_vision_info(messages)
+        inputs = self.processor(
+            text=[text], images=image_inputs, videos=video_inputs, padding=True, return_tensors="pt"
+        )
+        inputs = inputs.to(self.model.device)
+
+        generated_ids = self.model.generate(**inputs, max_new_tokens=max_new_tokens)
+        generated_ids_trimmed = [
+            out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
+        ]
+        return self.processor.batch_decode(
+            generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
+        )[0]
+
+
+# %%
+
+# Step 1: Prepare data
+if USE_HF_DATASET:
+    from datasets import load_dataset
+
+    dataset = load_dataset(DATASET_NAME, split=DATASET_SPLIT)
+    N = len(dataset) if MAX_DOCS is None else min(MAX_DOCS, len(dataset))
+    filepaths: list[str] = []
+    images: list[Image.Image] = []
+    for i in tqdm(range(N), desc="Loading dataset", total=N ):
+        p = dataset[i]
+        # Compose a descriptive identifier for printing later
+        identifier = f"arXiv:{p['paper_arxiv_id']}|title:{p['paper_title']}|page:{int(p['page_number'])}|id:{p['page_id']}"
+        print(identifier)
+        filepaths.append(identifier)
+        images.append(p["page_image"])  # PIL Image
+else:
+    _maybe_convert_pdf_to_images(PDF, PAGES_DIR)
+    filepaths, images = _load_images_from_dir(PAGES_DIR)
+    if not images:
+        raise RuntimeError(
+            f"No images found in {PAGES_DIR}. Provide PDF path in PDF variable or ensure images exist."
+        )
+
+
+# %%
+# Step 2: Load model and processor
+model_name, model, processor, device_str, device, dtype = _load_colvision(MODEL)
+print(f"Using model={model_name}, device={device_str}, dtype={dtype}")
+
+
+# %%
+
+# %%
+# Step 3: Build or load index
+retriever: Optional[LeannMultiVector] = None
+if not REBUILD_INDEX:
+    try:
+        one_vec = _embed_images(model, processor, [images[0]])[0]
+        retriever = _load_retriever_if_index_exists(INDEX_PATH, dim=int(one_vec.shape[-1]))
+    except Exception:
+        retriever = None
+
+if retriever is None:
+    doc_vecs = _embed_images(model, processor, images)
+    retriever = _build_index(INDEX_PATH, doc_vecs, filepaths)
+
+
+# %%
+# Step 4: Embed query and search
+q_vec = _embed_queries(model, processor, [QUERY])[0]
+results = retriever.search(q_vec.float().numpy(), topk=TOPK, first_stage_k=FIRST_STAGE_K)
+if not results:
+    print("No results found.")
+else:
+    print(f'Top {len(results)} results for query: "{QUERY}"')
+    top_images: list[Image.Image] = []
+    for rank, (score, doc_id) in enumerate(results, start=1):
+        path = filepaths[doc_id]
+        # For HF dataset, path is a descriptive identifier, not a real file path
+        print(f"{rank}) MaxSim: {score:.4f}, Page: {path}")
+        top_images.append(images[doc_id])
+
+    if SAVE_TOP_IMAGE:
+        from pathlib import Path as _Path
+
+        base = _Path(SAVE_TOP_IMAGE)
+        base.parent.mkdir(parents=True, exist_ok=True)
+        for rank, img in enumerate(top_images[:TOPK], start=1):
+            if base.suffix:
+                out_path = base.parent / f"{base.stem}_rank{rank}{base.suffix}"
+            else:
+                out_path = base / f"retrieved_page_rank{rank}.png"
+            img.save(str(out_path))
+            print(f"Saved retrieved page (rank {rank}) to: {out_path}")
+
+## TODO stange results of second page of DeepSeek-V2 rather than the first page
+
+# %%
+# Step 5: Similarity maps for top-K results
+if results and SIMILARITY_MAP:
+    token_idx = None if SIM_TOKEN_IDX < 0 else int(SIM_TOKEN_IDX)
+    from pathlib import Path as _Path
+
+    output_base = _Path(SIM_OUTPUT) if SIM_OUTPUT else None
+    for rank, img in enumerate(top_images[:TOPK], start=1):
+        if output_base:
+            if output_base.suffix:
+                out_dir = output_base.parent
+                out_name = f"{output_base.stem}_rank{rank}{output_base.suffix}"
+                out_path = str(out_dir / out_name)
+            else:
+                out_dir = output_base
+                out_dir.mkdir(parents=True, exist_ok=True)
+                out_path = str(out_dir / f"similarity_map_rank{rank}.png")
+        else:
+            out_path = None
+        chosen_idx, max_sim = _generate_similarity_map(
+            model=model,
+            processor=processor,
+            image=img,
+            query=QUERY,
+            token_idx=token_idx,
+            output_path=out_path,
+        )
+        if out_path:
+            print(
+                f"Saved similarity map for rank {rank}, token #{chosen_idx} (max={max_sim:.2f}) to: {out_path}"
+            )
+        else:
+            print(
+                f"Computed similarity map for rank {rank}, token #{chosen_idx} (max={max_sim:.2f})"
+            )
+
+
+# %%
+# Step 6: Optional answer generation
+if results and ANSWER:
+    qwen = QwenVL(device=device_str)
+    response = qwen.answer(QUERY, top_images[:TOPK], max_new_tokens=MAX_NEW_TOKENS)
+    print("\nAnswer:")
+    print(response)

apps/semantic_file_search/leann-plus-temporal-search.py

@@ -0,0 +1,183 @@
+#!/usr/bin/env python3
+import re
+import sys
+from datetime import datetime, timedelta
+from pathlib import Path
+
+from leann import LeannSearcher
+
+INDEX_PATH = str(Path("./").resolve() / "demo.leann")
+
+
+class TimeParser:
+    def __init__(self):
+        # Main pattern: captures optional fuzzy modifier, number, unit, and optional "ago"
+        self.pattern = r"(?:(around|about|roughly|approximately)\s+)?(\d+)\s+(hour|day|week|month|year)s?(?:\s+ago)?"
+
+        # Compile for performance
+        self.regex = re.compile(self.pattern, re.IGNORECASE)
+
+        # Stop words to remove before regex parsing
+        self.stop_words = {
+            "in",
+            "at",
+            "of",
+            "by",
+            "as",
+            "me",
+            "the",
+            "a",
+            "an",
+            "and",
+            "any",
+            "find",
+            "search",
+            "list",
+            "ago",
+            "back",
+            "past",
+            "earlier",
+        }
+
+    def clean_text(self, text):
+        """Remove stop words from text"""
+        words = text.split()
+        cleaned = " ".join(word for word in words if word.lower() not in self.stop_words)
+        return cleaned
+
+    def parse(self, text):
+        """Extract all time expressions from text"""
+        # Clean text first
+        cleaned_text = self.clean_text(text)
+
+        matches = []
+        for match in self.regex.finditer(cleaned_text):
+            fuzzy = match.group(1)  # "around", "about", etc.
+            number = int(match.group(2))
+            unit = match.group(3).lower()
+
+            matches.append(
+                {
+                    "full_match": match.group(0),
+                    "fuzzy": bool(fuzzy),
+                    "number": number,
+                    "unit": unit,
+                    "range": self.calculate_range(number, unit, bool(fuzzy)),
+                }
+            )
+
+        return matches
+
+    def calculate_range(self, number, unit, is_fuzzy):
+        """Convert to actual datetime range and return ISO format strings"""
+        units = {
+            "hour": timedelta(hours=number),
+            "day": timedelta(days=number),
+            "week": timedelta(weeks=number),
+            "month": timedelta(days=number * 30),
+            "year": timedelta(days=number * 365),
+        }
+
+        delta = units[unit]
+        now = datetime.now()
+        target = now - delta
+
+        if is_fuzzy:
+            buffer = delta * 0.2  # 20% buffer for fuzzy
+            start = (target - buffer).isoformat()
+            end = (target + buffer).isoformat()
+        else:
+            start = target.isoformat()
+            end = now.isoformat()
+
+        return (start, end)
+
+
+def search_files(query, top_k=15):
+    """Search the index and return results"""
+    # Parse time expressions
+    parser = TimeParser()
+    time_matches = parser.parse(query)
+
+    # Remove time expressions from query for semantic search
+    clean_query = query
+    if time_matches:
+        for match in time_matches:
+            clean_query = clean_query.replace(match["full_match"], "").strip()
+
+    # Check if clean_query is less than 4 characters
+    if len(clean_query) < 4:
+        print("Error: add more input for accurate results.")
+        return
+
+    # Single query to vector DB
+    searcher = LeannSearcher(INDEX_PATH)
+    results = searcher.search(
+        clean_query if clean_query else query, top_k=top_k, recompute_embeddings=False
+    )
+
+    # Filter by time if time expression found
+    if time_matches:
+        time_range = time_matches[0]["range"]  # Use first time expression
+        start_time, end_time = time_range
+
+        filtered_results = []
+        for result in results:
+            # Access metadata attribute directly (not .get())
+            metadata = result.metadata if hasattr(result, "metadata") else {}
+
+            if metadata:
+                # Check modification date first, fall back to creation date
+                date_str = metadata.get("modification_date") or metadata.get("creation_date")
+
+                if date_str:
+                    # Convert strings to datetime objects for proper comparison
+                    try:
+                        file_date = datetime.fromisoformat(date_str)
+                        start_dt = datetime.fromisoformat(start_time)
+                        end_dt = datetime.fromisoformat(end_time)
+
+                        # Compare dates properly
+                        if start_dt <= file_date <= end_dt:
+                            filtered_results.append(result)
+                    except (ValueError, TypeError):
+                        # Handle invalid date formats
+                        print(f"Warning: Invalid date format in metadata: {date_str}")
+                        continue
+
+        results = filtered_results
+
+    # Print results
+    print(f"\nSearch results for: '{query}'")
+    if time_matches:
+        print(
+            f"Time filter: {time_matches[0]['number']} {time_matches[0]['unit']}(s) {'(fuzzy)' if time_matches[0]['fuzzy'] else ''}"
+        )
+        print(
+            f"Date range: {time_matches[0]['range'][0][:10]} to {time_matches[0]['range'][1][:10]}"
+        )
+    print("-" * 80)
+
+    for i, result in enumerate(results, 1):
+        print(f"\n[{i}] Score: {result.score:.4f}")
+        print(f"Content: {result.text}")
+
+        # Show metadata if present
+        metadata = result.metadata if hasattr(result, "metadata") else None
+        if metadata:
+            if "creation_date" in metadata:
+                print(f"Created: {metadata['creation_date']}")
+            if "modification_date" in metadata:
+                print(f"Modified: {metadata['modification_date']}")
+        print("-" * 80)
+
+
+if __name__ == "__main__":
+    if len(sys.argv) < 2:
+        print('Usage: python search_index.py "<search query>" [top_k]')
+        sys.exit(1)
+
+    query = sys.argv[1]
+    top_k = int(sys.argv[2]) if len(sys.argv) > 2 else 15
+
+    search_files(query, top_k)

apps/semantic_file_search/leann_index_builder.py

@@ -0,0 +1,82 @@
+#!/usr/bin/env python3
+import json
+import sys
+from pathlib import Path
+
+from leann import LeannBuilder
+
+
+def process_json_items(json_file_path):
+    """Load and process JSON file with metadata items"""
+
+    with open(json_file_path, encoding="utf-8") as f:
+        items = json.load(f)
+
+    # Guard against empty JSON
+    if not items:
+        print("⚠️  No items found in the JSON file. Exiting gracefully.")
+        return
+
+    INDEX_PATH = str(Path("./").resolve() / "demo.leann")
+    builder = LeannBuilder(backend_name="hnsw", is_recompute=False)
+
+    total_items = len(items)
+    items_added = 0
+    print(f"Processing {total_items} items...")
+
+    for idx, item in enumerate(items):
+        try:
+            # Create embedding text sentence
+            embedding_text = f"{item.get('Name', 'unknown')} located at {item.get('Path', 'unknown')} and size {item.get('Size', 'unknown')} bytes with content type {item.get('ContentType', 'unknown')} and kind {item.get('Kind', 'unknown')}"
+
+            # Prepare metadata with dates
+            metadata = {}
+            if "CreationDate" in item:
+                metadata["creation_date"] = item["CreationDate"]
+            if "ContentChangeDate" in item:
+                metadata["modification_date"] = item["ContentChangeDate"]
+
+            # Add to builder
+            builder.add_text(embedding_text, metadata=metadata)
+            items_added += 1
+
+        except Exception as e:
+            print(f"\n⚠️  Warning: Failed to process item {idx}: {e}")
+            continue
+
+        # Show progress
+        progress = (idx + 1) / total_items * 100
+        sys.stdout.write(f"\rProgress: {idx + 1}/{total_items} ({progress:.1f}%)")
+        sys.stdout.flush()
+
+    print()  # New line after progress
+
+    # Guard against no successfully added items
+    if items_added == 0:
+        print("⚠️  No items were successfully added to the index. Exiting gracefully.")
+        return
+
+    print(f"\n✅ Successfully processed {items_added}/{total_items} items")
+    print("Building index...")
+
+    try:
+        builder.build_index(INDEX_PATH)
+        print(f"✓ Index saved to {INDEX_PATH}")
+    except ValueError as e:
+        if "No chunks added" in str(e):
+            print("⚠️  No chunks were added to the builder. Index not created.")
+        else:
+            raise
+
+
+if __name__ == "__main__":
+    if len(sys.argv) != 2:
+        print("Usage: python build_index.py <json_file>")
+        sys.exit(1)
+
+    json_file = sys.argv[1]
+    if not Path(json_file).exists():
+        print(f"Error: File {json_file} not found")
+        sys.exit(1)
+
+    process_json_items(json_file)

apps/semantic_file_search/spotlight_index_dump.py

@@ -0,0 +1,265 @@
+#!/usr/bin/env python3
+"""
+Spotlight Metadata Dumper for Vector DB
+Extracts only essential metadata for semantic search embeddings
+Output is optimized for vector database storage with minimal fields
+"""
+
+import json
+import sys
+from datetime import datetime
+
+# Check platform before importing macOS-specific modules
+if sys.platform != "darwin":
+    print("This script requires macOS (uses Spotlight)")
+    sys.exit(1)
+
+from Foundation import NSDate, NSMetadataQuery, NSPredicate, NSRunLoop
+
+# EDIT THIS LIST: Add or remove folders to search
+# Can be either:
+# - Folder names relative to home directory (e.g., "Desktop", "Downloads")
+# - Absolute paths (e.g., "/Applications", "/System/Library")
+SEARCH_FOLDERS = [
+    "Desktop",
+    "Downloads",
+    "Documents",
+    "Music",
+    "Pictures",
+    "Movies",
+    # "Library",  # Uncomment to include
+    # "/Applications",  # Absolute path example
+    # "Code/Projects",  # Subfolder example
+    # Add any other folders here
+]
+
+
+def convert_to_serializable(obj):
+    """Convert NS objects to Python serializable types"""
+    if obj is None:
+        return None
+
+    # Handle NSDate
+    if hasattr(obj, "timeIntervalSince1970"):
+        return datetime.fromtimestamp(obj.timeIntervalSince1970()).isoformat()
+
+    # Handle NSArray
+    if hasattr(obj, "count") and hasattr(obj, "objectAtIndex_"):
+        return [convert_to_serializable(obj.objectAtIndex_(i)) for i in range(obj.count())]
+
+    # Convert to string
+    try:
+        return str(obj)
+    except Exception:
+        return repr(obj)
+
+
+def dump_spotlight_data(max_items=10, output_file="spotlight_dump.json"):
+    """
+    Dump Spotlight data using public.item predicate
+    """
+    # Build full paths from SEARCH_FOLDERS
+    import os
+
+    home_dir = os.path.expanduser("~")
+    search_paths = []
+
+    print("Search locations:")
+    for folder in SEARCH_FOLDERS:
+        # Check if it's an absolute path or relative
+        if folder.startswith("/"):
+            full_path = folder
+        else:
+            full_path = os.path.join(home_dir, folder)
+
+        if os.path.exists(full_path):
+            search_paths.append(full_path)
+            print(f"  ✓ {full_path}")
+        else:
+            print(f"  ✗ {full_path} (not found)")
+
+    if not search_paths:
+        print("No valid search paths found!")
+        return []
+
+    print(f"\nDumping {max_items} items from Spotlight (public.item)...")
+
+    # Create query with public.item predicate
+    query = NSMetadataQuery.alloc().init()
+    predicate = NSPredicate.predicateWithFormat_("kMDItemContentTypeTree CONTAINS 'public.item'")
+    query.setPredicate_(predicate)
+
+    # Set search scopes to our specific folders
+    query.setSearchScopes_(search_paths)
+
+    print("Starting query...")
+    query.startQuery()
+
+    # Wait for gathering to complete
+    run_loop = NSRunLoop.currentRunLoop()
+    print("Gathering results...")
+
+    # Let it gather for a few seconds
+    for i in range(50):  # 5 seconds max
+        run_loop.runMode_beforeDate_(
+            "NSDefaultRunLoopMode", NSDate.dateWithTimeIntervalSinceNow_(0.1)
+        )
+        # Check gathering status periodically
+        if i % 10 == 0:
+            current_count = query.resultCount()
+            if current_count > 0:
+                print(f"  Found {current_count} items so far...")
+
+    # Continue while still gathering (up to 2 more seconds)
+    timeout = NSDate.dateWithTimeIntervalSinceNow_(2.0)
+    while query.isGathering() and timeout.timeIntervalSinceNow() > 0:
+        run_loop.runMode_beforeDate_(
+            "NSDefaultRunLoopMode", NSDate.dateWithTimeIntervalSinceNow_(0.1)
+        )
+
+    query.stopQuery()
+
+    total_results = query.resultCount()
+    print(f"Found {total_results} total items")
+
+    if total_results == 0:
+        print("No results found")
+        return []
+
+    # Process items
+    items_to_process = min(total_results, max_items)
+    results = []
+
+    # ONLY relevant attributes for vector embeddings
+    # These provide essential context for semantic search without bloat
+    attributes = [
+        "kMDItemPath",  # Full path for file retrieval
+        "kMDItemFSName",  # Filename for display & embedding
+        "kMDItemFSSize",  # Size for filtering/ranking
+        "kMDItemContentType",  # File type for categorization
+        "kMDItemKind",  # Human-readable type for embedding
+        "kMDItemFSCreationDate",  # Temporal context
+        "kMDItemFSContentChangeDate",  # Recency for ranking
+    ]
+
+    print(f"Processing {items_to_process} items...")
+
+    for i in range(items_to_process):
+        try:
+            item = query.resultAtIndex_(i)
+            metadata = {}
+
+            # Extract ONLY the relevant attributes
+            for attr in attributes:
+                try:
+                    value = item.valueForAttribute_(attr)
+                    if value is not None:
+                        # Keep the attribute name clean (remove kMDItem prefix for cleaner JSON)
+                        clean_key = attr.replace("kMDItem", "").replace("FS", "")
+                        metadata[clean_key] = convert_to_serializable(value)
+                except (AttributeError, ValueError, TypeError):
+                    continue
+
+            # Only add if we have at least a path
+            if metadata.get("Path"):
+                results.append(metadata)
+
+        except Exception as e:
+            print(f"Error processing item {i}: {e}")
+            continue
+
+    # Save to JSON
+    with open(output_file, "w", encoding="utf-8") as f:
+        json.dump(results, f, indent=2, ensure_ascii=False)
+
+    print(f"\n✓ Saved {len(results)} items to {output_file}")
+
+    # Show summary
+    print("\nSample items:")
+    import os
+
+    home_dir = os.path.expanduser("~")
+
+    for i, item in enumerate(results[:3]):
+        print(f"\n[Item {i + 1}]")
+        print(f"  Path: {item.get('Path', 'N/A')}")
+        print(f"  Name: {item.get('Name', 'N/A')}")
+        print(f"  Type: {item.get('ContentType', 'N/A')}")
+        print(f"  Kind: {item.get('Kind', 'N/A')}")
+
+        # Handle size properly
+        size = item.get("Size")
+        if size:
+            try:
+                size_int = int(size)
+                if size_int > 1024 * 1024:
+                    print(f"  Size: {size_int / (1024 * 1024):.2f} MB")
+                elif size_int > 1024:
+                    print(f"  Size: {size_int / 1024:.2f} KB")
+                else:
+                    print(f"  Size: {size_int} bytes")
+            except (ValueError, TypeError):
+                print(f"  Size: {size}")
+
+        # Show dates
+        if "CreationDate" in item:
+            print(f"  Created: {item['CreationDate']}")
+        if "ContentChangeDate" in item:
+            print(f"  Modified: {item['ContentChangeDate']}")
+
+    # Count by type
+    type_counts = {}
+    for item in results:
+        content_type = item.get("ContentType", "unknown")
+        type_counts[content_type] = type_counts.get(content_type, 0) + 1
+
+    print(f"\nTotal items saved: {len(results)}")
+
+    if type_counts:
+        print("\nTop content types:")
+        for ct, count in sorted(type_counts.items(), key=lambda x: x[1], reverse=True)[:5]:
+            print(f"  {ct}: {count} items")
+
+    # Count by folder
+    folder_counts = {}
+    for item in results:
+        path = item.get("Path", "")
+        for folder in SEARCH_FOLDERS:
+            # Build the full folder path
+            if folder.startswith("/"):
+                folder_path = folder
+            else:
+                folder_path = os.path.join(home_dir, folder)
+
+            if path.startswith(folder_path):
+                folder_counts[folder] = folder_counts.get(folder, 0) + 1
+                break
+
+    if folder_counts:
+        print("\nItems by location:")
+        for folder, count in sorted(folder_counts.items(), key=lambda x: x[1], reverse=True):
+            print(f"  {folder}: {count} items")
+
+    return results
+
+
+def main():
+    # Parse arguments
+    if len(sys.argv) > 1:
+        try:
+            max_items = int(sys.argv[1])
+        except ValueError:
+            print("Usage: python spot.py [number_of_items]")
+            print("Default: 10 items")
+            sys.exit(1)
+    else:
+        max_items = 10
+
+    output_file = sys.argv[2] if len(sys.argv) > 2 else "spotlight_dump.json"
+
+    # Run dump
+    dump_spotlight_data(max_items=max_items, output_file=output_file)
+
+
+if __name__ == "__main__":
+    main()

benchmarks/bm25_diskann_baselines/README.md

@@ -0,0 +1,23 @@
+BM25 vs DiskANN Baselines
+
+```bash
+aws s3 sync s3://powerrag-diskann-rpj-wiki-20250824-224037-194d640c/bm25_rpj_wiki/index_en_only/ benchmarks/data/indices/bm25_index/
+aws s3 sync s3://powerrag-diskann-rpj-wiki-20250824-224037-194d640c/diskann_rpj_wiki/ benchmarks/data/indices/diskann_rpj_wiki/
+```
+
+- Dataset: `benchmarks/data/queries/nq_open.jsonl` (Natural Questions)
+- Machine-specific; results measured locally with the current repo.
+
+DiskANN (NQ queries, search-only)
+- Command: `uv run --script benchmarks/bm25_diskann_baselines/run_diskann.py`
+- Settings: `recompute_embeddings=False`, embeddings precomputed (excluded from timing), batching off, caching off (`cache_mechanism=2`, `num_nodes_to_cache=0`)
+- Result: avg 0.011093 s/query, QPS 90.15 (p50 0.010731 s, p95 0.015000 s)
+
+BM25
+- Command: `uv run --script benchmarks/bm25_diskann_baselines/run_bm25.py`
+- Settings: `k=10`, `k1=0.9`, `b=0.4`, queries=100
+- Result: avg 0.028589 s/query, QPS 34.97 (p50 0.026060 s, p90 0.043695 s, p95 0.053260 s, p99 0.055257 s)
+
+Notes
+- DiskANN measures search-only latency on real NQ queries (embeddings computed beforehand and excluded from timing).
+- Use `benchmarks/bm25_diskann_baselines/run_diskann.py` for DiskANN; `benchmarks/bm25_diskann_baselines/run_bm25.py` for BM25.

benchmarks/bm25_diskann_baselines/run_bm25.py

@@ -0,0 +1,183 @@
+# /// script
+# dependencies = [
+#   "pyserini"
+# ]
+# ///
+# sudo pacman -S jdk21-openjdk
+# export JAVA_HOME=/usr/lib/jvm/java-21-openjdk
+# sudo archlinux-java status
+# sudo archlinux-java set java-21-openjdk
+# set -Ux JAVA_HOME /usr/lib/jvm/java-21-openjdk
+# fish_add_path --global $JAVA_HOME/bin
+# set -Ux LD_LIBRARY_PATH $JAVA_HOME/lib/server $LD_LIBRARY_PATH
+# which javac # Should be /usr/lib/jvm/java-21-openjdk/bin/javac
+
+import argparse
+import json
+import os
+import sys
+import time
+from statistics import mean
+
+
+def load_queries(path: str, limit: int | None) -> list[str]:
+    queries: list[str] = []
+    # Try JSONL with a 'query' or 'text' field; fallback to plain text (one query per line)
+    _, ext = os.path.splitext(path)
+    if ext.lower() in {".jsonl", ".json"}:
+        with open(path, encoding="utf-8") as f:
+            for line in f:
+                line = line.strip()
+                if not line:
+                    continue
+                try:
+                    obj = json.loads(line)
+                except json.JSONDecodeError:
+                    # Not strict JSONL? treat the whole line as the query
+                    queries.append(line)
+                    continue
+                q = obj.get("query") or obj.get("text") or obj.get("question")
+                if q:
+                    queries.append(str(q))
+    else:
+        with open(path, encoding="utf-8") as f:
+            for line in f:
+                s = line.strip()
+                if s:
+                    queries.append(s)
+
+    if limit is not None and limit > 0:
+        queries = queries[:limit]
+    return queries
+
+
+def percentile(values: list[float], p: float) -> float:
+    if not values:
+        return 0.0
+    s = sorted(values)
+    k = (len(s) - 1) * (p / 100.0)
+    f = int(k)
+    c = min(f + 1, len(s) - 1)
+    if f == c:
+        return s[f]
+    return s[f] + (s[c] - s[f]) * (k - f)
+
+
+def main():
+    ap = argparse.ArgumentParser(description="Standalone BM25 latency benchmark (Pyserini)")
+    ap.add_argument(
+        "--bm25-index",
+        default="benchmarks/data/indices/bm25_index",
+        help="Path to Pyserini Lucene index directory",
+    )
+    ap.add_argument(
+        "--queries",
+        default="benchmarks/data/queries/nq_open.jsonl",
+        help="Path to queries file (JSONL with 'query'/'text' or plain txt one-per-line)",
+    )
+    ap.add_argument("--k", type=int, default=10, help="Top-k to retrieve (default: 10)")
+    ap.add_argument("--k1", type=float, default=0.9, help="BM25 k1 (default: 0.9)")
+    ap.add_argument("--b", type=float, default=0.4, help="BM25 b (default: 0.4)")
+    ap.add_argument("--limit", type=int, default=100, help="Max queries to run (default: 100)")
+    ap.add_argument(
+        "--warmup", type=int, default=5, help="Warmup queries not counted in latency (default: 5)"
+    )
+    ap.add_argument(
+        "--fetch-docs", action="store_true", help="Also fetch doc contents (slower; default: off)"
+    )
+    ap.add_argument("--report", type=str, default=None, help="Optional JSON report path")
+    args = ap.parse_args()
+
+    try:
+        from pyserini.search.lucene import LuceneSearcher
+    except Exception:
+        print("Pyserini not found. Install with: pip install pyserini", file=sys.stderr)
+        raise
+
+    if not os.path.isdir(args.bm25_index):
+        print(f"Index directory not found: {args.bm25_index}", file=sys.stderr)
+        sys.exit(1)
+
+    queries = load_queries(args.queries, args.limit)
+    if not queries:
+        print("No queries loaded.", file=sys.stderr)
+        sys.exit(1)
+
+    print(f"Loaded {len(queries)} queries from {args.queries}")
+    print(f"Opening BM25 index: {args.bm25_index}")
+    searcher = LuceneSearcher(args.bm25_index)
+    # Some builds of pyserini require explicit set_bm25; others ignore
+    try:
+        searcher.set_bm25(k1=args.k1, b=args.b)
+    except Exception:
+        pass
+
+    latencies: list[float] = []
+    total_searches = 0
+
+    # Warmup
+    for i in range(min(args.warmup, len(queries))):
+        _ = searcher.search(queries[i], k=args.k)
+
+    t0 = time.time()
+    for i, q in enumerate(queries):
+        t1 = time.time()
+        hits = searcher.search(q, k=args.k)
+        t2 = time.time()
+        latencies.append(t2 - t1)
+        total_searches += 1
+
+        if args.fetch_docs:
+            # Optional doc fetch to include I/O time
+            for h in hits:
+                try:
+                    _ = searcher.doc(h.docid)
+                except Exception:
+                    pass
+
+        if (i + 1) % 50 == 0:
+            print(f"Processed {i + 1}/{len(queries)} queries")
+
+    t1 = time.time()
+    total_time = t1 - t0
+
+    if latencies:
+        avg = mean(latencies)
+        p50 = percentile(latencies, 50)
+        p90 = percentile(latencies, 90)
+        p95 = percentile(latencies, 95)
+        p99 = percentile(latencies, 99)
+        qps = total_searches / total_time if total_time > 0 else 0.0
+    else:
+        avg = p50 = p90 = p95 = p99 = qps = 0.0
+
+    print("BM25 Latency Report")
+    print(f"  queries: {total_searches}")
+    print(f"  k: {args.k}, k1: {args.k1}, b: {args.b}")
+    print(f"  avg per query: {avg:.6f} s")
+    print(f"  p50/p90/p95/p99: {p50:.6f}/{p90:.6f}/{p95:.6f}/{p99:.6f} s")
+    print(f"  total time: {total_time:.3f} s, qps: {qps:.2f}")
+
+    if args.report:
+        payload = {
+            "queries": total_searches,
+            "k": args.k,
+            "k1": args.k1,
+            "b": args.b,
+            "avg_s": avg,
+            "p50_s": p50,
+            "p90_s": p90,
+            "p95_s": p95,
+            "p99_s": p99,
+            "total_time_s": total_time,
+            "qps": qps,
+            "index_dir": os.path.abspath(args.bm25_index),
+            "fetch_docs": bool(args.fetch_docs),
+        }
+        with open(args.report, "w", encoding="utf-8") as f:
+            json.dump(payload, f, indent=2)
+        print(f"Saved report to {args.report}")
+
+
+if __name__ == "__main__":
+    main()

benchmarks/bm25_diskann_baselines/run_diskann.py

@@ -0,0 +1,124 @@
+# /// script
+# dependencies = [
+#   "leann-backend-diskann"
+# ]
+# ///
+
+import argparse
+import json
+import time
+from pathlib import Path
+
+import numpy as np
+
+
+def load_queries(path: Path, limit: int | None) -> list[str]:
+    out: list[str] = []
+    with open(path, encoding="utf-8") as f:
+        for line in f:
+            obj = json.loads(line)
+            out.append(obj["query"])
+            if limit and len(out) >= limit:
+                break
+    return out
+
+
+def main() -> None:
+    ap = argparse.ArgumentParser(
+        description="DiskANN baseline on real NQ queries (search-only timing)"
+    )
+    ap.add_argument(
+        "--index-dir",
+        default="benchmarks/data/indices/diskann_rpj_wiki",
+        help="Directory containing DiskANN files",
+    )
+    ap.add_argument("--index-prefix", default="ann")
+    ap.add_argument("--queries-file", default="benchmarks/data/queries/nq_open.jsonl")
+    ap.add_argument("--num-queries", type=int, default=200)
+    ap.add_argument("--top-k", type=int, default=10)
+    ap.add_argument("--complexity", type=int, default=62)
+    ap.add_argument("--threads", type=int, default=1)
+    ap.add_argument("--beam-width", type=int, default=1)
+    ap.add_argument("--cache-mechanism", type=int, default=2)
+    ap.add_argument("--num-nodes-to-cache", type=int, default=0)
+    args = ap.parse_args()
+
+    index_dir = Path(args.index_dir).resolve()
+    if not index_dir.is_dir():
+        raise SystemExit(f"Index dir not found: {index_dir}")
+
+    qpath = Path(args.queries_file).resolve()
+    if not qpath.exists():
+        raise SystemExit(f"Queries file not found: {qpath}")
+
+    queries = load_queries(qpath, args.num_queries)
+    print(f"Loaded {len(queries)} queries from {qpath}")
+
+    # Compute embeddings once (exclude from timing)
+    from leann.api import compute_embeddings as _compute
+
+    embs = _compute(
+        queries,
+        model_name="facebook/contriever-msmarco",
+        mode="sentence-transformers",
+        use_server=False,
+    ).astype(np.float32)
+    if embs.ndim != 2:
+        raise SystemExit("Embedding compute failed or returned wrong shape")
+
+    # Build searcher
+    from leann_backend_diskann.diskann_backend import DiskannSearcher as _DiskannSearcher
+
+    index_prefix_path = str(index_dir / args.index_prefix)
+    searcher = _DiskannSearcher(
+        index_prefix_path,
+        num_threads=int(args.threads),
+        cache_mechanism=int(args.cache_mechanism),
+        num_nodes_to_cache=int(args.num_nodes_to_cache),
+    )
+
+    # Warmup (not timed)
+    _ = searcher.search(
+        embs[0:1],
+        top_k=args.top_k,
+        complexity=args.complexity,
+        beam_width=args.beam_width,
+        prune_ratio=0.0,
+        recompute_embeddings=False,
+        batch_recompute=False,
+        dedup_node_dis=False,
+    )
+
+    # Timed loop
+    times: list[float] = []
+    for i in range(embs.shape[0]):
+        t0 = time.time()
+        _ = searcher.search(
+            embs[i : i + 1],
+            top_k=args.top_k,
+            complexity=args.complexity,
+            beam_width=args.beam_width,
+            prune_ratio=0.0,
+            recompute_embeddings=False,
+            batch_recompute=False,
+            dedup_node_dis=False,
+        )
+        times.append(time.time() - t0)
+
+    times_sorted = sorted(times)
+    avg = float(sum(times) / len(times))
+    p50 = times_sorted[len(times) // 2]
+    p95 = times_sorted[max(0, int(len(times) * 0.95) - 1)]
+
+    print("\nDiskANN (NQ, search-only) Report")
+    print(f"  queries: {len(times)}")
+    print(
+        f"  k: {args.top_k}, complexity: {args.complexity}, beam_width: {args.beam_width}, threads: {args.threads}"
+    )
+    print(f"  avg per query: {avg:.6f} s")
+    print(f"  p50/p95: {p50:.6f}/{p95:.6f} s")
+    print(f"  QPS: {1.0 / avg:.2f}")
+
+
+if __name__ == "__main__":
+    main()

benchmarks/data/prompts_g5/prompt_dump_hotpot_hnsw.txt

@@ -1,484 +0,0 @@
-=== Prompt Dump for HOTPOT + HNSW ===
-Total prompts: 50
-Showing first 20 prompts:
-
-==================================================
-PROMPT #1:
-==================================================
-Le samedi est le 272 jour de l'année 1962. Naissances Janne Andersson, footballeur suédois devenu entraîneur Lauren Brice (morte le ), actrice pornographique américaine Néstor Clausen, footballeur argentin Nicky Walker, footballeur britannique Olivier Trassard, footballeur français reconverti entraîneur Roger Bart, acteur américain Solal, chanteur français Stefan Niederseer, skieur alpin autrichien Décès Owe Jonsson (né le ), athlète suédois spécialiste du sprint Robert Burton (né le ), acteur américain Sunsiaré de Larcône (née le ), auteure française Événements investiture d'Ahmed Benbella à la présidence de l'Assemblée nationale d'Algérie. lancement de Alouette 1 dans l'espace. Le Canada devient le troisième pays à avoir un satellite artificiel autour de la terre Voir aussi 29 septembre et septembre 1962Le samedi est le 136 jour de l'année 1959. Naissances Andrew Litton, chef d'orchestre et pianiste américain Cynthia del Águila, femme politique guatémaltèque Gerard Donovan, écrivain irlandais Greg Johnston, rameur néo-zélandais Ioánnis Andrianós, homme politique grec Jean-Michel Veillon, musicien français Mare Winningham, actrice américaine Martin Pado, homme politique slovaque Mitch Webster, joueur américain de baseball Ofelia Cano, actrice mexicaine Roberto Piazza, acteur, chanteur, écrivain et styliste argentin Vincent de Brus, réalisateur, scénariste et perchman français Yannick Letty, écrivain français Décès Edward Winslow Gifford (né le ), ethnologue américain Elisha Scott (né le ), joueur de football irlandais Friedrich Grimm (né le ), personnalité politique allemande Gyula Kiss (né le ), joueur international hongrois de football William Hammond Wright (né le ), astronome américain Événements Fin du championnat d'Israël de football 1958-1959 Début du tour d'Italie de 1959Derrick Williams, basketteur américain. 1992 : Yasser al-Shahrani, footballeur saoudien. (38 sélections en équipe nationale). Elena Cecchini, cycliste sur piste et sur route italienne. Championne du monde de cyclisme sur route du contre la montre par équipes 2018. Jón Daði Böðvarsson, footballeur islandais. (26 sélections en équipe nationale). 1993 : Pierre Lees-Melou, footballeur français. Danijel Petković, footballeur monténégrin. (17 sélections en équipe nationale). Norman Powell, basketteur américain. 1994 : Matt Murray, hockeyeur sur glace canadien. Alexandra Raisman, gymnaste artistique américaine. Championne olympique du concours général par équipes et du sol, médaillée de bronze de la poutre aux Jeux de Londres 2012 puis championne olympique du concours général par équipes et médaillée d'argent du concours général individuel ainsi qu'au sol aux Jeux de Rio 2016. Championne du monde de gymnastique artistique du concours général par équipes 2011 et 2015. Hannah Blundell footballeuse anglaise 1995 : José Gayà, footballeur espagnol. 1996 : Dorian Godon, cycliste sur route français. David Pastrňák, hockeyeur sur glace tchèque. 1998 : Javi Puado, footballeur espagnol. 1999 : Ibrahima Konaté, footballeur français. Terem Moffi, footballeur nigérian. 2000 : Krzysztof Kubica, footballeur polonais. Claire Liu, joueuse de tennis américaine. Arthur Theate, footballeur belge. (4 sélections en équipe nationale). 2001 : Douaa Foudali, escrimeuse marocaine. Sturla Ottesen, footballeur norvégien. 2002 : Manfred Ugalde, footballeur costaricien. (1 sélection en équipe nationale). 2004 : Pedro Acosta, pilote de moto espagnol. 2005 : Bella Sims, nageuse américaine. Médaillée d'argent du relais 4 × 200 m nage libre aux Jeux olympiques de Tokyo en 2021. Décès de 1901 à 1950
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: 2014 S/S is the debut album of a South Korean boy group that was formed by who?
-A: YG Entertainment
-
-Q: What government position was held by the woman who portrayed Corliss Archer in the film Kiss and Tell?
-A: Chief of Protocol
-
-Q: What science fantasy young adult series, told in first person, has a set of companion books narrating the stories of enslaved worlds and alien species?
-A: Animorphs
-
-Q: Are the Laleli Mosque and Esma Sultan Mansion located in the same neighborhood?
-A: no
-
-Q: The director of the romantic comedy "Big Stone Gap" is based in what New York city?
-A: Greenwich Village, New York City
-
-Q: Were Scott Derrickson and Ed Wood of the same nationality?
-A: 
-==================================================
-
-==================================================
-PROMPT #2:
-==================================================
-Whiting); Mildred's irritating younger brother Raymond (Tommy Bernard, Kenny Godkin); and Corliss's rival, Betty Cameron (Delores Crane). Meet Corliss Archer was written by F. Hugh Herbert, who first introduced the character and her friends in a Good Housekeeping magazine story "A Private Affair," the first of a series of stories. Kiss and Tell was a 1943 play that was adapted for a 1945 film starring Shirley Temple. The 1949 sequel, A Kiss for Corliss, was re-released in 1954. Television Herbert's Meet Corliss Archer property was adapted to television in 1951 and 1952 with live broadcasts produced and aired by CBS. In 1954 and 1955 Ziv Television Programs produced a syndicated telefilm series starring Ann Baker and Mary Brian. One of the show's unique features was the occasional cut to a comic-book-style drawing, with announcer's commentary, that illustrated the current story situation and was used several times during each episode. The program was produced by Ziv Productions. Several episodes of the Ziv version are available on DVD, and some are also available from the Internet Archive. Robin Morgan portrayed Corliss in a live telecast of Kiss and Tell on The Alcoa Hour (August 5, 1956), with Warren Berlinger as Dexter. Comics Radio listeners had to use their imaginations to visualize Corliss, her friends and her town. But those imaginations got a boost in 1948 when the Meet Corliss Archer comic book, published by Fox Feature Syndicate, came out in three issues from March to July 1948. Al Feldstein (Albert B. Feldstein), later the editor of Mad, wasEdward G. Robinson's Big Town, The Eddie Bracken Show, Favorite Story, Four Star Playhouse, The Gallant Heart, One Man's Family, Sears Radio Theater, and Stars over Hollywood. She also played several characters including Joanne Allen and Mrs. Hodges on the radio show Adventures in Odyssey produced by Focus on the Family broadcasting. She co-starred with Jimmy Lydon in the CBS situation comedy Young Love (1949–50), and she had recurring roles on The Adventures of Ozzie and Harriet (as teenager Emmy Lou), The Red Skelton Show, and People Are Funny. She recorded with jazz vocalist Mel Torme and his vocal group the Mel-Tones. Her eight-year run starring as teenager Corliss Archer on CBS's Meet Corliss Archer left a lasting impression, though Shirley Temple starred in the film adaptations, Kiss and Tell and A Kiss for Corliss. The radio program was the CBS answer to NBC's popular A Date with Judy. Despite the long run of Meet Corliss Archer, fewer than 24 episodes are known to exist. Waldo later turned down the offer to portray Corliss in a television adaptation (the role was recast with Ann Baker taking the role for TV). In 1948, the Meet Corliss Archer comic book, using Waldo's likeness, published by Fox Feature Syndicate, appeared for a run of three issues from March to July 1948, using the original scripts. On April 1, 1948, Waldo married playwright Robert E. Lee, the writing partner of Jerome Lawrence. The couple had two children, and remained married until his death in 1994. Waldo made a rare on-screenKiss and Tell is a 1945 American comedy film starring then 17-year-old Shirley Temple as Corliss Archer. In the film, two teenage girls cause their respective parents much concern when they start to become interested in boys. The parents' bickering about which girl is the worse influence causes more problems than it solves. The movie was based on the Broadway play Kiss and Tell, which was based on the Corliss Archer short stories. The stories, play and movie were all written by F. Hugh Herbert. A sequel film, A Kiss for Corliss, was released in 1949 and also starred Temple, but was not written by Herbert. Plot When their booth at a USO bazaar fails to attract customers, teenager Corliss Archer suggests to her best friend, Mildred Pringle, that they sell kisses. The idea becomes a success among the soldiers visiting the bazaar, and business is booming, until the girls' mothers find out about it. Despite the fact that it was Corliss' idea, Mrs. Archer blames Mildred for the girls' behavior. Mrs. Archer's assumptions greatly upset Mr. Pringle when he hears about them at the Pringle dinner table. He decides the family, including Mildred and their son Raymond, will refrain from socializing with the Archers in the future. That same night, Corliss pretends to be older than she is and starts flirting shamelessly with young Private Jimmy Earhart, who has been invited into the Archer home for dinner. Corliss is actually dating the boy next door, Dexter Franklin. Late that night, Lenny Archer returns home on
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Were Scott Derrickson and Ed Wood of the same nationality?
-A: yes
-
-Q: 2014 S/S is the debut album of a South Korean boy group that was formed by who?
-A: YG Entertainment
-
-Q: What science fantasy young adult series, told in first person, has a set of companion books narrating the stories of enslaved worlds and alien species?
-A: Animorphs
-
-Q: Are the Laleli Mosque and Esma Sultan Mansion located in the same neighborhood?
-A: no
-
-Q: The director of the romantic comedy "Big Stone Gap" is based in what New York city?
-A: Greenwich Village, New York City
-
-Q: What government position was held by the woman who portrayed Corliss Archer in the film Kiss and Tell?
-A: 
-==================================================
-
-==================================================
-PROMPT #3:
-==================================================
-this from happening, and is joined by one of the descendants of the 'altered' humans from the second book. Unknown to Lalelelang, she will discover shocking things about not only the telepathic humans, but a couple of alien species from the first two books as well as herself. Significant parts of the action of the novels take place between the various non-human species that make up both The Weave and the Amplitur's allies, giving an outside perspective of "normal" human interactions and events as the author's apparent commentary on current events. Reception Kirkus Reviews criticized the first book as having no plot or action. Publishers Weekly praises the second book in the series as being an improvement over the first and having more depth. References Science fiction book series Science fiction novel trilogies Novels by Alan Dean FosterTV science fiction series: Star Trek, Stargate, Firefly, Battlestar Galactica and Farscape (the void of the series title being both an allusion to this, as well as 'Sliding Void' being used by spacers as slang for sub-warp transit). The first novel, Void all the Way Down, focuses on the inclusion of Calder, an exiled prince from a medieval-level failed colony world, into the ship's crew as they embark on running supplies to an illegal mining operation. The second novel, Anomalous Thrust, finds the crew stranded on a world boycotted by civilisation for its human colonists' enslavement of the local sentient species. Here, they find they can only escape by taking part in a solar sail race to recover valuable alien garbage jettisoned by a moon-sized vessel making a periodic tour of multiple galaxies - a trope inspired by Rendezvous with Rama. Far-called series The Far-called sequence is a fantasy series from Stephen Hunt set on the world of Pellas. The first book in the series, In Dark Service was published by Gollancz, the science fiction and fantasy imprint of Hachette and Orion Publishing, on 15 May 2014. The second novel in the series, Foul Tide's Turning, was released on 21 May 2015. The third novel in the series, The Stealers' War, was released on 16 March 2016. Hunt mentioned at Comic-Con that his books set in Pellas draw on the events of the U.S. Civil War, just as Game of Thrones draw on the feuding of 15th-century England's Wars of the Roses (an idea that cametale of forbidden love and Machiavellian politics. With detailed history, architecture, customs, inventions, and landscape, Jae creates a fascinating new world to explore." The School Library Journal called it "An engaging and amusing read", but said "the concepts and characters are not fully developed." Study Breaks listed the first book as one of "5 YA Books By Hispanic Authors To Read During National Hispanic Heritage Month", and said "it starts big and stays strong throughout." Meanwhile, Kirkus Reviews called the first book "deeply flawed", and complained of "pacing issues, plot holes, and flat secondary characters", but said the second book was "an improvement on the first outing" and that the third book was "dramatically improved over previous entries". The books' alien vocabulary has been both praised and criticized. The School Library Journal indicated that the language was part of what made the books' world "beautifully conceived...with unique details", but Kirkus Reviews complained about "the inaccessible and forced jargon". References Young adult novel series Science fiction novel trilogies LGBT speculative fiction novels Debut science fiction novels Book series introduced in 2016 Novels set on fictional planets LGBT-related young adult novels 2010s LGBT novels Works published under a pseudonym American science fiction novels American LGBT novels American young adult novels 2016 American novels 2017 American novels 2018 American novels 2018 debut novels Skyhorse Publishing books 2016 LGBT-related literary works
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Were Scott Derrickson and Ed Wood of the same nationality?
-A: yes
-
-Q: What government position was held by the woman who portrayed Corliss Archer in the film Kiss and Tell?
-A: Chief of Protocol
-
-Q: 2014 S/S is the debut album of a South Korean boy group that was formed by who?
-A: YG Entertainment
-
-Q: Are the Laleli Mosque and Esma Sultan Mansion located in the same neighborhood?
-A: no
-
-Q: The director of the romantic comedy "Big Stone Gap" is based in what New York city?
-A: Greenwich Village, New York City
-
-Q: What science fantasy young adult series, told in first person, has a set of companion books narrating the stories of enslaved worlds and alien species?
-A: 
-==================================================
-
-==================================================
-PROMPT #4:
-==================================================
-Laleli Mosque () is a mosque in the Abdi Çavuş quarter of the walled city of Nicosia, currently located in North Nicosia. It is located on Ali Ruhi Street. Its name, meaning "the Mosque with Tulips", is thought to be derived from the tulip motifs adorning its original minaret. Originally a small medieval chapel, it was enlarged and converted to a mosque in the 19th century. History It is understood from the mosque's architecture that the building was originally a medieval chapel. After the Ottoman conquest, it was initially converted to a small mosque without a minaret (mescit), and was known as "Laleli Cami-i Minareli Mescidi" in the 18th century. In the late 20th century, a marble inscription dated to 1742-43 was found in a building at the southwestern corner of the mosque's courtyard. Written in the talik style by a calligrapher named "Ârif", the inscription revealed the existence of a teacher's school ("Laleli Cami-i Minareli Mescit Muallimhâne Mektebi") at the site at the time. In 1827, it was enlarged by the addition of a minaret, minbar, mihrab, a new ceiling and a narthex (son cemaat yeri) and thus converted into a full mosque. The construction was financed by Ali Ruhi Efendi, the governor of Cyprus at the time, and the mosque was inscribed amongst the property of the newly founded Ali Ruhi Efendi Foundation on 24 December 1827 for maintenance. The responsibility for the maintenance changed hands over the course of the 19th century, as it was recorded amongst the property of Seyit Mehmet AğaLaleli (meaning 'with tulips' in Turkish) is a neighbourhood of Fatih, Istanbul, Turkey, lying between Beyazıt and Aksaray. It is known for its large textile wholesaling business and is home to the Literature and Science Faculties of Istanbul University, designed by Sedad Hakkı Eldem and Emin Onat in the 1940s. It is served by a stop on the T1 tram line which runs along Ordu Caddesi. The most prominent historic monument in Laleli is the Laleli Mosque, a work of architect Mehmed Tahir Ağa that was originally constructed in the 1760s. It was built for Sultan Mustafa III whose tomb it contains. An attractive sebil or water dispensary stands on the street side of the complex surrounding the mosque. The mosque stands above a large basement that is now filled with clothes shops. Across the road from the mosque is the Koca Ragıp Paşa complex, also designed by Mehmed Tahir Ağa in 1762. It was undergoing restoration for much of the 2010s. Lurking in the back streets is the much older Bodrum Mosque (AKA Mesih Paşa Cami), which started life as a 10th-century Byzantine church attached to the Myreiaion Palace. Beside it is an underground cistern, probably of similar date. Both stand on the site of a lost Rotunda dating back to the fifth century which is believed to have been the second largest such circular Roman temple after the Pantheon in Rome itself. Also in Laleli is the Big Stone Han (Büyük Taş Hanı in Turkish) which was probably part of the Laleli Mosque complexThe Esma Sultan Mansion (), a historical yalı located on the Bosphorus in the Ortaköy neighborhood of Istanbul, Turkey and named after its original owner Princess Esma Sultan, is used today as a cultural center after being redeveloped. History The three-storey brick manor was designed by the renowned architect Sarkis Balyan and finished in 1875 next to Ortaköy Mosque. It was presented to the Princess Esma Sultan, the daughter of Ottoman Sultan Abdulaziz, as a wedding gift in 1889. The mansion remained in the possession of the Ottoman dynasty until 1915. Subsequently, the building was used first as a tobacco warehouse and then as a coal depot from 1920 until 1975 when it was destroyed by a fire. Redevelopment The ruin, consisting only of the outer walls of the building, was purchased in the early 1990s by The Marmara Collection. Following a renovation with additions designed by architects Haluk Sezgin and Philippe Robert, the mansion was opened in 2001 as a multipurpose event venue. Within the brick exterior, which was left as the original, a steel and glass structure is incorporated. According to the GAD Architecture Group, a redesign by the architect Gökhan Avcıoğlu was completed in 2005. The building includes a bar, a restaurant, and an event hall at several levels. The building is situated in a garden of . The ground floor is wide, long and high. The first floor is merged with the second floor, having dimensions of wide, long and high. Esma Sultana Mansion, run by The Marmara Hotel chain as a
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Were Scott Derrickson and Ed Wood of the same nationality?
-A: yes
-
-Q: What government position was held by the woman who portrayed Corliss Archer in the film Kiss and Tell?
-A: Chief of Protocol
-
-Q: What science fantasy young adult series, told in first person, has a set of companion books narrating the stories of enslaved worlds and alien species?
-A: Animorphs
-
-Q: 2014 S/S is the debut album of a South Korean boy group that was formed by who?
-A: YG Entertainment
-
-Q: The director of the romantic comedy "Big Stone Gap" is based in what New York city?
-A: Greenwich Village, New York City
-
-Q: Are the Laleli Mosque and Esma Sultan Mansion located in the same neighborhood?
-A: 
-==================================================
-
-==================================================
-PROMPT #5:
-==================================================
-of Virginia at Wise (2001), and The University of New Haven, Connecticut (2005, 2016). She is host of the Library of Virginia Literary Awards (11 years), was host of the Poets and Writers Gala in New York City (2016), The Audio Publishers Association Audies Gala (2014), and the Barnes & Noble Writers for Writers Awards. She is permanent host of the Erma Bombeck/Arizona Women's Board Annual Authors Luncheon that is committed to the prevention of kidney disease through awareness, education and research. Novels Big Stone Gap: a novel. New York: Random House, 2000. Big Cherry Holler: a Big Stone Gap novel. New York: Random House, 2001 Milk Glass Moon : a Big Stone Gap novel. New York: Random House, 2002. Lucia, Lucia: a novel. New York: Random House, 2004. Home to Big Stone Gap; a novel. New York: Random House, 2006. The Queen of the Big Time : a novel. New York: Random House, 2004. Rococo: a novel. New York: Random House, 2005. Very Valentine: a novel. New York: Random House, 2009. Viola in reel life. New York: Harper Teen, 2009. Brava, Valentine: a novel. Harper Collins Publishers, 2010 Viola In the Spotlight New York: Harper Teen, 2011. The Shoemaker's Wife: a novel.New York: Harper Collins Publishers, 2012. The Supreme Macaroni Company : a novel. New York : Harper Collins Publishers, 2013. All The Stars in the Heavens : a novel. New York: Harper, Harper Collins Publishers, 2015. Kiss Carlo. New York: Harper Collins Publishers, 2017 Tony's Wife: a novel. New York: Harper Collins Publishers, 2018of Adriana Trigiani. During the 1990s, Trigiani wrote and directed an award-winning documentary Queen of The Big Time (1996), the story of her father's hometown of Roseto, Pennsylvania, shown in film festivals in London and Hong Kong, co-produced Green Chimneys, and later contributed to PBS documentary The Italian Americans. In 2014, Trigiani directed the major motion picture Big Stone Gap (film), a romantic comedy film adaptation of her namesake bestselling novel, produced by Donna Gigliotti for Altar Identity Studios, a subsidiary of Media Society. Big Stone Gap (film) is a story of family secrets and self-discovery in an Appalachian coal-mining town of the late 1970s. The award-winning ensemble cast includes Ashley Judd, Whoopi Goldberg, Jane Krakowski, Jenna Elfman, and Patrick Wilson. Released on October 9, 2015 by Picturehouse (company), Trigiani's narrative directorial debut arrived nearly 30 years after the sale of her first screenplay, Three to Get Married, produced by Kate Benton in 1986. Opening the Virginia Film Festival, Big Stone Gap (film) was ranked among the top 250 grossing women directed films of 2014. Media appearances Trigiani and her work have regularly been featured on NBC's Today Show. She was profiled on CBS Sunday Morning, appeared on The View, Good Day NY with Rosanna Scotto and Lori Stokes, and is heard regularly on NPR around the country. Trigiani has lectured at New York University and the New School for Social Research, has been a commencement speaker and received honorary degrees from Emory & Henry College (2018), Saint Mary's College, Notre Dame, Indiana (2002), The UniversityBig Stone Gap is a 2014 American drama romantic comedy film written and directed by Adriana Trigiani and produced by Donna Gigliotti for Altar Identity Studios, a subsidiary of Media Society. Based on Trigiani's 2000 best-selling novel of the same name, the story is set in the actual Virginia town of Big Stone Gap circa 1970s. The film had its world premiere at the Virginia Film Festival on November 6, 2014. The film was released on October 9, 2015, by Picturehouse. The film was released in Blu-Ray by Universal Pictures Home Entertainment on February 2, 2016. Plot In 1978, 40-year-old independent woman Ave Maria Mulligan owns her dead father's pharmacy in her hometown of Big Stone Gap, Virginia. Ave's mother, an immigrant from Italy, is the town seamstress. Ave is heavily involved in her community, home delivering medications to the country folk, volunteering on the coal mining town's Emergency Response Team, and directing the town's annual production of "Trail of the Lonesome Pine", based on the novel by John Fox Jr. Ave delivers medications to Nan McChesney and is pleased to see Nan's son Jack, a local coal miner and a former schoolmate. Jack tries to make small talk, but Ave leaves when Nan tells her that 40 is not too old to still have children. Spec Broadwater, the town's lawyer and Ave's friend, brings news that her momma has died. At the graveside Fleeta Mullins, Ave's outspoken friend and pharmacy worker, chastises Ave for placing a rose on her father's grave, remarking how badly he
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Were Scott Derrickson and Ed Wood of the same nationality?
-A: yes
-
-Q: What government position was held by the woman who portrayed Corliss Archer in the film Kiss and Tell?
-A: Chief of Protocol
-
-Q: What science fantasy young adult series, told in first person, has a set of companion books narrating the stories of enslaved worlds and alien species?
-A: Animorphs
-
-Q: Are the Laleli Mosque and Esma Sultan Mansion located in the same neighborhood?
-A: no
-
-Q: 2014 S/S is the debut album of a South Korean boy group that was formed by who?
-A: YG Entertainment
-
-Q: The director of the romantic comedy "Big Stone Gap" is based in what New York city?
-A: 
-==================================================
-
-==================================================
-PROMPT #6:
-==================================================
-SS501 is a South Korean boy band consisting of five members: Kim Hyun-joong, Heo Young-saeng, Kim Kyu-jong, Park Jung-min and Kim Hyung-jun. They debuted in 2005 with the EP Warning, which included their first number one song, "Never Again". In 2007, SS501 entered the Japanese market with the maxi single Kokoro, which debuted at number five on the Oricon chart. The following year, in 2008, they became the first South Korean artists to win the award for Best New Artist at the Japan Gold Disc Awards. SS501 has released one studio album and seven EPs in Korean, and two studio albums and three EPs in Japanese. The group has been inactive since 2010, however the sub-group Double S 301, composed of members Heo Young-saeng, Kim Kyu-jong and Kim Hyung-jun, continues to release music. Etymology The name of the band is a combination of letters and numbers that have special meaning. The first “S” stands for "superstar", the second “S” is an abbreviation for "singer", and the combination of 5, 0, and 1 symbolizes "five members united as one forever". History Pre-debut Before debuting with SS501, most of the members had some experience in the entertainment world. Kim Hyun-joong was initially about to debut in a five-member group which included Han Yeon of B2Y (now disbanded) who was to be the leader and Kim would be the youngest member. During 2001 and 2002, while working as a waiter in a family restaurant located at Jamsil-dong, Kim was introduced to a new set up management company CEO. WhileS (에스) is a South Korean project group consisting of three members: Kangta, Lee Ji-hoon and Shin Hye-sung. The group debuted in 2003, under the SM Entertainment label. After 11 years, they released and promoted another mini-album in 2014. History In 2003, S released their first album Fr.In.Cl, which stands for Friends in Classic. In 2014, after more than a decade in hiatus, the group released their second mini-album Autumn Breeze on October 27. As member Kangta wrote and produced the songs, it was said that his "unique music style and the trio’s harmonious voices will create beautiful ballads". On October 18, the group performed their title track "Without You (하고 싶은 거 다)" for the first time at SM Entertainment's agency-wide concert SM Town in Shanghai. The music video for the song was released on October 24, starring Kwon Yuri of Girls' Generation. The group continued to promote the song on various South Korean music programs, such as on Immortal Songs 2 on November 3. Band members Kangta Shin Hye-sung Lee Ji-hoon Discography Studio albums EP Concert tours SMTown 2014: SM Town Live World Tour IV Awards and nominations References South Korean boy bands South Korean contemporary R&B musical groups South Korean musical trios Musical groups established in 2003 Musical groups disestablished in 2014 Kangta2014 S/S is the debut album of South Korean group Winner. It was released on August 12, 2014, by the group's record label, YG Entertainment. The members were credited for writing the lyrics and composing the majority of the album's songs. Composition The members produced the majority of the music for this album themselves, with the help of other producers such as Choice 37, B.I., Airplay, and others. The album was highlighted for incorporating elements generally absent from K-pop releases, including hints of acoustic and alternative rock. Track listing Reception The title track "Empty" became a hit in South Korea, topping the Gaon Chart and Billboard's K-Pop Hot 100 chart. Internationally, 2014 S/S took the top spot in Billboard's World Album Chart. Charts Sales References External links 2014 albums YG Entertainment albums Winner (band) albums Korean-language albums
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Were Scott Derrickson and Ed Wood of the same nationality?
-A: yes
-
-Q: What government position was held by the woman who portrayed Corliss Archer in the film Kiss and Tell?
-A: Chief of Protocol
-
-Q: What science fantasy young adult series, told in first person, has a set of companion books narrating the stories of enslaved worlds and alien species?
-A: Animorphs
-
-Q: Are the Laleli Mosque and Esma Sultan Mansion located in the same neighborhood?
-A: no
-
-Q: The director of the romantic comedy "Big Stone Gap" is based in what New York city?
-A: Greenwich Village, New York City
-
-Q: 2014 S/S is the debut album of a South Korean boy group that was formed by who?
-A: 
-==================================================
-
-==================================================
-PROMPT #7:
-==================================================
-Donald G. Malcolm (March 26, 1919 - June 18, 2007) was an American organizational theorist, professor and dean at Cal State L.A.'s College of Business and Economics and management consultant, known as co-developer of the Performance, Evaluation, and Review Technique (PERT). Biography Malcolm was born and raised in Indianapolis. In 1940 he obtained his BSc in Public Service Engineering from Purdue University, and after the war his MSc in industrial engineering. In World War II Malcolm served on an aircraft carrier in the Pacific as radioman. After his graduation Malcolm started his academic career as research team leader of the Operations Research Team at the Johns Hopkins University. In 1950 he was lecturer in Mechanical Engineering, and by 1952 assistant professor of industrial engineering at the University of California, Berkeley. In the 1950s he was also management consultant for Booz, Allen and Hamilton in Chicago, Illinois, and in those days participated with the US Navy Special Projects Office, especially with Willard Fazar, in the development of PERT. In 1954-1955 he was president of the American Institute of Industrial Engineers, now Institute of Industrial and Systems Engineers. In 1962 he settled as management consultant with his own consultancy firm Management Technology Inc. From 1962 to 1967 he also directed the National Safety Standards division of the US Department of Transportation. From 1970 he was one of the Senior Vice Presidents of Computer Applications, Inc., and from 1972 to 1981 he served as dean of the College of Business and Economics at the California State University, Los Angeles.William James Reddin also known as Bill Reddin (May 10, 1930 – June 20, 1999) was a British-born management behavioralist, theorist, writer, and consultant. His published works examined and explained how managers in profit and non-profit organizations behaved under certain situations and conditions. The focus of his work was to understand to what extent managers were effective in their role and successful in managing situations to have the right impact on the organization's objectives. Through extensive research Reddin concluded that there is no ideal management style. He put forward that there was only one realistic and unambiguous definition of managerial effectiveness, the extent to which a manager or leader achieves the output requirements of the position. This is the manager's or leader's only job: to be effective. Reddin was often quoted as saying both in his writings, to his clients and to his students, that there is no ideal style of managing; and there is no one way to make an organization more effective. He wrote in his 1988 book, The Output Oriented Manager, "… no list exists in the world showing characteristics of effective managers, or, of effective organizations, which apply generally. He went on to write that his works' intent were to serve as a substitute for prescriptive management-guru advice prevalent in modern business, to enable the manager and leader in diagnosing what is the true situation and what are the true needs. They served managers and leaders to make sound decisions on how best to arrive at their planned objectives. This concept ofThis is an annotated list of important business writers. It is in alphabetical order based on last name. For quick navigation, click on one of the letters: A David Aaker (born 1938) - marketing, brand strategy Wil van der Aalst James Abegglen (1926–2007) - management and business in Japan Bodo Abel Russell L. Ackoff (1919–2009) - operations research, organizational theory John Adair (born 1934) - leadership Karol Adamiecki (1866–1933) - management Ichak Adizes Niclas Adler (born 1971) - Swedish organizational theorist Charles Constance César Joseph Matthieu d'Agoult Yoji Akao Ali Akdemir Mark Albion (born 1951) - values-based business Howard E. Aldrich (born 1940s) - American sociologist and organizational theorist Leon P. Alford (1877–1942) - scientific management Tim Ambler (born 1938) - marketing effectiveness Igor Ansoff (1918–2002) - strategic management Ingeman Arbnor Chris Argyris (1923–2013) - learning systems, learning organization Horace Lucian Arnold (1837–1915) Neal Ashkanasy B Stephen R. Barley (born 1953) - technology, organizational change, organizational culture Chester Barnard (1886–1961) - management Gary S. Becker Charles Bedaux (1886–1944) - scientific management Warren Bennis (1925–2014) - leadership studies Per Olof Berg (born 1946) - Swedish organizational theorist Manfred Berliner Björn Bjerke Patrick Blackett, Baron Blackett (1897–1974) - operations research Ken Blanchard Charles Bosanquet Matthew Boulton (1728–1809) - labor productivity Marvin Bower Richard Boyatzis (born 1946) - emotional intelligence, behavior change, and competence Leland Lawrence Briggs (1893–1975) - American accounting scholar John Seely Brown Wilfred Brown, Baron Brown Nils Brunsson (born 1946) - institutionalized hypLine truncated
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Were Scott Derrickson and Ed Wood of the same nationality?
-A: yes
-
-Q: What government position was held by the woman who portrayed Corliss Archer in the film Kiss and Tell?
-A: Chief of Protocol
-
-Q: What science fantasy young adult series, told in first person, has a set of companion books narrating the stories of enslaved worlds and alien species?
-A: Animorphs
-
-Q: Are the Laleli Mosque and Esma Sultan Mansion located in the same neighborhood?
-A: no
-
-Q: The director of the romantic comedy "Big Stone Gap" is based in what New York city?
-A: Greenwich Village, New York City
-
-Q: Who was known by his stage name Aladin and helped organizations improve their performance as a consultant?
-A: 
-==================================================
-
-==================================================
-PROMPT #8:
-==================================================
-Civic Center). The Lewiston Maineiacs, the only American team to have played in the Quebec Major Junior Hockey League, played their first season in 2003–2004 and dissolved the team after the 2010–2011 season. The Colisée is also the home to the state Class A and Class B high school hockey championships each year. The city as a whole is known for its strong passion for the game of hockey, likely related to its French American heritage. Two Lewiston schools, Lewiston High School and St. Dominic Regional High School (now in Auburn), combine for over half of the state class A high school hockey championships in the state's history. During the 2013–2014 American Hockey League season, the Portland Pirates played their first 12 home games at the Colisée while the Cumberland County Civic Center is being renovated. The junior Maine Nordiques of the North American Hockey League have played their home games at the Colisée since 2019. Ali vs. Liston rematch In May 1965, Lewiston became the venue for a heavyweight title rematch between Muhammad Ali and Sonny Liston; Ali had defeated Liston in a controversial fight in Miami Beach, Florida in February 1964, and the World Boxing Council was demanding an immediate rematch, which was against World Boxing Association rules (the WBA eventually stripped Ali of their title). The rematch was originally planned to be held in Boston, but was halted by Massachusetts boxing authorities due to licensing issues. Promoters were eventually able to frame a lucrative deal moving the fight to the Colisée in Lewiston.As the venue held less than 3,700 spectators, only 2,434 fans were present, setting an all-time record for the lowest attendance for a heavyweight championship match. The fight was the scene of the famous photograph of Ali standing over Liston taunting him with his glove. Ali won by first round knockout. Lewiston Twins (1891–1930) Lewiston was home to minor league baseball. Beginning in 1891, Lewiston was home to the Lewiston Twins and other teams, who played in various seasons through 1930. Lewiston teams played as members of the New England League (1891–1896, 1901), Maine State League (1907), Atlantic Association (1908) and New England League (1914–1915, 1919, 1926–1930). Baseball Hall of Fame member Jesse Burkett managed the Lewiston Twins in 1928 and 1929. Between 1901 and 1919, Lewiston teams played home games at A.A.A. Park. Beginning in 1926, the Lewiston Twins played home games at Lewiston Athletic Park. Maine Nordiques (1973–1977) The Maine Nordiques were a professional hockey team that operated in the former North American Hockey League from 1973 to 1977. They were based at the Central Maine Youth Center in Lewiston. The Nordiques served as a farm club for the Quebec Nordiques of the World Hockey Association. Lewiston Maineiacs The Lewiston Maineiacs were a major junior hockey team that played in the Quebec Major Junior Hockey League (QMJHL). The Maineiacs moved to Lewiston in 2003 from Sherbrooke, Quebec, and were the only team in the QMJHL in the United States. They played their home games at the Androscoggin Bank Colisée. In 2006–2007, the Maineiacs wonThe Lewiston Maineiacs were a junior ice hockey team of the Quebec Major Junior Hockey League based in Lewiston, Maine, United States. The team played its home games at the Androscoggin Bank Colisée. They were the second QMJHL team in the United States, and the only one to play a full season. They won the President's Cup in 2007, becoming the only American based franchise to accomplish the feat. On May 31, 2011, the QMJHL decided to fold the team following eight years of financial losses. History Their franchise was granted for the 1969–70 season. They played in Trois-Rivières, Quebec, from 1969 to 1992. They were known as the Ducs from 1969 to 1973, and as the Draveurs from 1973 to 1992. They moved to Sherbrooke, Quebec, in 1992 to become the Faucons (1992–1998) and the Castors (1998–2003). The Castors moved to Lewiston, Maine, in 2003. The MAINEiacs clinched the Jean Rougeau Trophy for first overall in the regular season for 2006–07. During the same season, Lewiston increased its annual attendance totals to 94,903 tickets sold. In the playoffs that year, Lewiston went 16–1 to cruise to their first President's Cup. 2007 Memorial Cup After winning the 2007 President's Cup, the MAINEiacs won eligibility to play in the 2007 Memorial Cup in Vancouver, British Columbia. Following an opening win over the Medicine Hat Tigers, the MAINEiacs suffered a huge loss with the injury of their captain, Marc-André Cliche, and never really found their scoring touch. They crashed out of the tournament in fourth place after losing
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Were Scott Derrickson and Ed Wood of the same nationality?
-A: yes
-
-Q: What government position was held by the woman who portrayed Corliss Archer in the film Kiss and Tell?
-A: Chief of Protocol
-
-Q: What science fantasy young adult series, told in first person, has a set of companion books narrating the stories of enslaved worlds and alien species?
-A: Animorphs
-
-Q: Are the Laleli Mosque and Esma Sultan Mansion located in the same neighborhood?
-A: no
-
-Q: The director of the romantic comedy "Big Stone Gap" is based in what New York city?
-A: Greenwich Village, New York City
-
-Q: The arena where the Lewiston Maineiacs played their home games can seat how many people?
-A: 
-==================================================
-
-==================================================
-PROMPT #9:
-==================================================
-recent film was the 2006 vampire horror film The Thirst, where she played a young girl with hemophilia named Sara. She graduated from East Ascension High School in Gonzales in 2005, and started her first year of college in 2006. She also plays guitar. She studied music at Tulane University in New Orleans, Louisiana, and also occasionally performs in regional theatre. From October 6 to October 16, 2011, Morton starred alongside McArdle in the new musical Greenwood at the New York Musical Theatre Festival. Morton then returned to Gonzales and taught drama at Ascension Christian High School. She now resides in Gonzales and works as a veterinarian technician in nearby Prairieville, Louisiana. Soundtracks Annie (1999 film soundtrack) Filmography References External links Alicia Morton on Instagram 1987 births Living people American child actresses Actresses from Louisiana American musical theatre actresses People from Gonzales, Louisiana 21st-century American womenAlicia Morton (born April 29, 1987) is an American former actress, singer, dancer, drama teacher, and veterinarian technician. She starred as Annie Bennett Warbucks in the 1999 Wonderful World of Disney production of Annie, which was based on the Broadway musical of the same name. Biography Morton was born and raised in Gonzales, Louisiana, and started singing when she was 18 months old. When she was 12, she played the titular role in Annie at her school in nearby Baton Rouge, Louisiana. She was 7 when she landed the part of young Cosette in Broadway's Les Misérables in 1996. Co-star Ricky Martin would rub makeup on her hands for good luck. In 1998, Morton beat 3,000 girls to win the part of Annie in the Disney television film Annie, which premiered on The Wonderful World of Disney on November 7, 1999, following a private screening at the New Amsterdam Theatre on November 1, 1999. In Annie, Morton co-starred with Lalaine, who would go on to star in the Disney Channel original series Lizzie McGuire from 2001 to 2004, and in the 2003 Disney Channel original movie You Wish!. In this role, Morton struggled with some of the dramatic moments. "The emotional scenes were hard for me," she says, until an acting coach suggested that she draw on her father's death. Her best friend, Alexis Kalehoff, is the daughter of Broadway's original Annie, Andrea McArdle. McArdle also had a cameo role in Morton's Annie, playing the Star-to-Be who has a solo in the song “N.Y.C.”. Her mostAnnie Morton (born October 8, 1970) is an American model born in Pennsylvania. She has appeared on the covers of British Vogue, ID, Marie Claire, and other magazines. She has been photographed by Helmut Newton; Peter Lindbergh; Annie Leibovitz; Richard Avedon; Juergen Teller; Paul Jasmin, Mary Ellen Mark, Stephen Shames, and Terry Richardson, and modeled for Donna Karan, Givenchy, Guerlain, Chanel, Harper's Bazaar, Sports Illustrated and Victoria's Secret. A long time vegetarian, an advocate for organic lifestyle choices and natural healthcare. She co-founded Tsi-La Organics, a "Green Luxury" company that creates and sells vegan, organic perfume and skin care products. Modeling She has appeared on many magazine covers and has been featured in several professional photobooks, Peter Lindbergh Selected Work, the cover of Juergen Teller By Juergen Teller, Helmut Newton Pages From The Glossies, and Donna Karan's book A Journey Of A Woman 20 Years. In 1997 she posed for the Pirelli Calendar with Richard Avedon. Morton has also appeared in several music videos. Death in Vegas' "Aisha" directed by Terry Richardson and The Rolling Stones' "Love So Strong" directed by David Fincher. Other business ventures Morton Co-founded Tsi-La Organics with her sister-in-law Natalie Morton in 2006. Tsi-La Organics, an award-winning "Green Luxury" company that creates and sells vegan, organic perfume and skin care products. Natalie and Annie created their "Green Luxury" Tsi~La product line as a supposedly healthy alternative to a predominantly synthetic market. Morton and her husband Michael Morpurgo of Dandelion own 3 School of Rock music schools (in Philadelphia, Pa, Doylestown, Pa and
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Were Scott Derrickson and Ed Wood of the same nationality?
-A: yes
-
-Q: What government position was held by the woman who portrayed Corliss Archer in the film Kiss and Tell?
-A: Chief of Protocol
-
-Q: What science fantasy young adult series, told in first person, has a set of companion books narrating the stories of enslaved worlds and alien species?
-A: Animorphs
-
-Q: Are the Laleli Mosque and Esma Sultan Mansion located in the same neighborhood?
-A: no
-
-Q: The director of the romantic comedy "Big Stone Gap" is based in what New York city?
-A: Greenwich Village, New York City
-
-Q: Who is older, Annie Morton or Terry Richardson?
-A: 
-==================================================
-
-==================================================
-PROMPT #10:
-==================================================
-the paintings came from Bolivia. The United States District Court for the Eastern District of Virginia concluded that, either way, Claimant still needed to provide documentation authorizing import of the paintings because both Peru and Bolivia are State Parties to the UNESCO Convention and both have import restriction agreements with the United States. Because Claimant could not provide such documentation, the paintings were subject to forfeiture. United States v. Twenty-Nine Pre-Columbian & Colonial Artifacts From Peru In 2010, United States Customs and Border Protection (CBP) seized Pre-Columbian and Colonial textiles, metals, lithics, and perishable remains from Peru at Miami International Airport. Claimant moved to dismiss for lack of subject matter jurisdiction, failure to state a claim, and denial of due process of law. First, the United States District Court for the Southern District of Florida found that the court did have subject matter jurisdiction. The court rejected Claimant's argument that the United States Court of International Trade had exclusive jurisdiction on the action. Second, the court found that the United States sufficiently stated a claim by showing the imported objects were protected material under the CCPIA. Third, the court rejected the claimant's argument that he was denied due process during CBP seizure proceedings because the CCPIA authorizes 90 days of artifact detention, and claimant did not protest the seizure despite notice and an opportunity to protest. Ancient Coin Collectors Guild cases The Ancient Coin Collectors Guild purchased twenty-two coins from a numismatic dealer in London in April 2009. The collection included twelve Chinese coins and sevenneed to be two law standards - the Pennsylvania law would apply to Piper and Scottish law would apply to Hartzell. A trial involving two sets of laws would be confusing to the jury. A lack of familiarity with Scottish law would also be confusing. Another powerful reason why Pennsylvania is a bad venue is that Scotland has a very strong interest in this litigation. There is "a local interest in having localized controversies decided at home." "Scotland has a very strong interest in this litigation. The accident occurred in its airspace. All of the decedents were Scottish. Apart from Piper and Hartzell, all potential plaintiffs and defendants are either Scottish or English. As we stated in Gilbert, there is 'a local interest in having localized controversies decided at home.' 330 U.S., at 509. Respondent argues that American citizens have an interest in ensuring that American manufacturers are deterred from producing defective products, and that additional deterrence might be obtained if Piper and Hartzell were tried in the United States, where they could be sued on the basis of both negligence and strict liability. However, the incremental deterrence that would be gained if this trial were held in an American court is likely to be insignificant. The American interest in this accident is simply not sufficient to justify the enormous commitment of judicial time and resources that would inevitably be required if the case were to be tried here." Disposition Reversed Rule If the remedy provided by the alternative forum is so clearly inadequate or unsatisfactoryforeign replacements. The United States Chamber of Commerce maintains a list of years when the random selection process (lottery) was implemented. The lottery process was challenged in Tenrec v. USCIS, a class action lawsuit in Oregon, but the case was decided against the plaintiffs. On June 28, 2021, the lottery process was challenged again in LIU et al. v. MAYORKAS et al., a lawsuit filed in United States District Court for the District of Columbia by 500+ FY 2022 H-1B applicants who didn't get selected in March 2021. Tax status of an individual with H-1B status The taxation of income for an individual with H-1B status depends on whether they are categorized as either nonresident aliens or resident aliens for tax purposes. A nonresident alien for tax purposes is only taxed on income that is effectively connected with a trade or business in the United States and United States-source income that is fixed, determinable, annual, or periodical. A resident alien for tax purposes is taxed on all income, including income from outside the United States. The classification is determined based on the substantial presence test. If the substantial presence test indicates that the individual is a resident, then income taxation is like any other U.S. person and may be filed using Form 1040 and the necessary schedules. Otherwise, the individual must file as a non-resident alien using Form 1040NR or Form 1040NR-EZ; the individual may claim a benefit from tax treaties that exist between the United States and the individual's country of citizenship. An individual in
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Were Scott Derrickson and Ed Wood of the same nationality?
-A: yes
-
-Q: What government position was held by the woman who portrayed Corliss Archer in the film Kiss and Tell?
-A: Chief of Protocol
-
-Q: What science fantasy young adult series, told in first person, has a set of companion books narrating the stories of enslaved worlds and alien species?
-A: Animorphs
-
-Q: Are the Laleli Mosque and Esma Sultan Mansion located in the same neighborhood?
-A: no
-
-Q: The director of the romantic comedy "Big Stone Gap" is based in what New York city?
-A: Greenwich Village, New York City
-
-Q: Are Local H and For Against both from the United States?
-A: 
-==================================================
-
-==================================================
-PROMPT #11:
-==================================================
-game, after the singing of "The Star-Spangled Banner", it is a tradition to sing the alma mater, "Crimson and the Blue". While singing the alma mater, it is tradition for the students in attendance to put an arm over the shoulders of their neighbors and slowly sway side to side, lifting their arms over their heads as the last line of the song is sung. The song is concluded by the Rock Chalk Chant. Post-game, the band will play a rendition of the Kansas state song "Home on the Range", which the crowd will stand up for similar to the national anthem. After singing The Star-Spangled Banner, while the opposing team is being introduced, members of the student section take out a copy of the student-run newspaper, The University Daily Kansan, and wave the paper in front of their faces, pretending to be reading it in an effort to show lack of interest in the opposing team. After the opponents are introduced, a short video is shown, detailing the history and the accomplishments of Kansas basketball. As the Jayhawks are introduced, the students rip up their newspapers and throw the confetti pieces of paper in the air as celebration. Whatever confetti remains is typically thrown in the air after the first basket made by the Jayhawks. If an opposing player fouls out of the game, the fans will "wave the wheat", waving their arms back and forth, as a sarcastic good-bye to the disqualified player, to the tune of "You Didn't Have Your Wheaties", from aI'm a Jayhawk is the fight song of the University of Kansas. History The original lyrics and music were written by George "Dumpy" Bowles in 1911. The lyrics to the song are generally not sung, instead, fans do a specific clap as the KU band plays the melody. The university bands do not play the intro, but only play the chorus of the song. The song was updated in 1958 to reflect the teams of the Big 8 Conference. The lyrics were rewritten again in 2010 to account for the departure of Nebraska and Colorado from the Big 12 Conference. The school has elected to not rewrite the lyrics following the departure of Texas A&M and Missouri from the league. Lyrics External links Birth of the Songs References University of Kansas American college songs College fight songs in the United States Big 12 Conference fight songs 1912 songs"Kansas Song (We’re From Kansas)" is a fight song of the University of Kansas. History The song was written by Carson J. Robison and Jack Riley. It was copyrighted 1922 by Abdallah Temple A.A.O.N.M.S. according to the sheet music. The university bands usually play only one verse of the song. References External links Lyrics and MP3s University of Kansas American college songs College fight songs in the United States Big 12 Conference fight songs 1922 songs Songs written by Carson Robison Songs about Kansas
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Were Scott Derrickson and Ed Wood of the same nationality?
-A: yes
-
-Q: What government position was held by the woman who portrayed Corliss Archer in the film Kiss and Tell?
-A: Chief of Protocol
-
-Q: What science fantasy young adult series, told in first person, has a set of companion books narrating the stories of enslaved worlds and alien species?
-A: Animorphs
-
-Q: Are the Laleli Mosque and Esma Sultan Mansion located in the same neighborhood?
-A: no
-
-Q: The director of the romantic comedy "Big Stone Gap" is based in what New York city?
-A: Greenwich Village, New York City
-
-Q: What is the name of the fight song of the university whose main campus is in Lawrence, Kansas and whose branch campuses are in the Kansas City metropolitan area?
-A: 
-==================================================
-
-==================================================
-PROMPT #12:
-==================================================
-David Diamond is an American screenwriter. He frequently collaborates with David Weissman. Early life Diamond was born on April 19, 1965, in Philadelphia, Pennsylvania. Diamond and Weissman met in high school, at Akiba Hebrew Academy (now Jack M. Barrack Hebrew Academy). They graduated in 1983. Career The duo sold their first spec script, The Whiz Kid, to 20th Century Fox in 1994. Their first produced project came in 2000, The Family Man starring Nicolas Cage and Tea Leoni. Cage's production company, Saturn Films, helped produce the film. The Family Man opened at #3 at the North American box office making $15.1 million in its opening weekend, behind What Women Want and Cast Away, which opened at the top spot. Diamond and Weissman next wrote the sci-fi comedy Evolution, helmed by Ivan Reitman. Evolution was based on a story by Don Jakoby, who turned it into a screenplay along with Diamond and Weissman. The film was originally written as a serious horror science fiction film, until director Reitman re-wrote much of the script. A short-lived animated series, Alienators: Evolution Continues, loosely based on the film, was broadcast months after the film was released. They partnered with Wedding Crashers producer Andrew Panay on Old Dogs and When In Rome. From 2018 to 2021, Diamond taught screenwriting at the Shalhevet High School in Los Angeles until moving back to Philadelphia. In 2019, the duo published a screenwriting manual, Bulletproof: Writing Scripts That Don't Get Shot Down. The book draws from their extensive film experience. References External links Living peopleRenard as Frank Mathieu Goldfeld as Victor Pablo-Noé Étienne as Lucas Critical response Evolution received generally positive reviews from critics. Review aggregator website Rotten Tomatoes reports an approval rating of 82% based on 62 reviews, with an average rating of 7.26/10. The site's critics' consensus reads: "Creepy, provocative, and aesthetically absorbing, Evolution marks a satisfying step forward for director/co-writer Lucile Hadzihalilovic." At Metacritic, which assigns a normalized rating out of 100 to reviews from mainstream critics, the film received an average score of 77 based on 19 reviews, indicating "generally favorable reviews". Accolades References External links 2015 films 2015 horror thriller films 2010s French-language films French horror thriller films Spanish horror thriller films Belgian horror thriller films Films directed by Lucile Hadžihalilović Films scored by Zacarías M. de la Riva 2010s French filmsDavid Weissman is an American screenwriter and director, most known for his comedies. He frequently collaborates with David Diamond. Diamond and Weissman met in high school, at Akiba Hebrew Academy (now Jack M. Barrack Hebrew Academy). They graduated in 1983. His film credits include Dream a Little Dream 2 (his only produced project not co-written with Diamond), The Family Man, Old Dogs, When in Rome, Evolution and the television film Minutemen. Career The duo sold their first spec script, The Whiz Kid, to 20th Century Fox in 1994. Their first produced project came in 2000, The Family Man starring Nicolas Cage and Tea Leoni. Cage's production company, Saturn Films, helped produce the film. The Family Man opened at #3 at the North American box office making $15.1 million in its opening weekend, behind What Women Want and Cast Away, which opened at the top spot. Diamond and Weissman next wrote the sci-fi comedy Evolution, helmed by Ivan Reitman. Evolution was based on a story by Don Jakoby, who turned it into a screenplay along with Diamond and Weissman. The film was originally written as a serious horror science fiction film, until director Reitman re-wrote much of the script. A short-lived animated series, Alienators: Evolution Continues, loosely based on the film, was broadcast months after the film was released. In 2005, they partnered with Wedding Crashers producer Andrew Panay on Old Dogs and When In Rome. In 2019, the duo published a screenwriting manual, Bulletproof: Writing Scripts That Don't Get Shot Down. The book draws from their
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Were Scott Derrickson and Ed Wood of the same nationality?
-A: yes
-
-Q: What government position was held by the woman who portrayed Corliss Archer in the film Kiss and Tell?
-A: Chief of Protocol
-
-Q: What science fantasy young adult series, told in first person, has a set of companion books narrating the stories of enslaved worlds and alien species?
-A: Animorphs
-
-Q: Are the Laleli Mosque and Esma Sultan Mansion located in the same neighborhood?
-A: no
-
-Q: The director of the romantic comedy "Big Stone Gap" is based in what New York city?
-A: Greenwich Village, New York City
-
-Q: What screenwriter with credits for "Evolution" co-wrote a film starring Nicolas Cage and Téa Leoni?
-A: 
-==================================================
-
-==================================================
-PROMPT #13:
-==================================================
-DJ-je egy klasszikus rockzenét játszó adónak a K-DST-nek a Grand Theft Auto: San Andreas nevű videójátékban. Rose feltűnt egy háttérszerepben a Guns N’ Roses akkori felállásával az 1988-as Clint Eastwood-filmben a Holtbiztos tippben. 2009-ben szóba került, hogy Rose lesz a producere annak a filmnek, ami az Almost Home című regényből készül. Diszkográfia Hollywood Rose Guns N’ Roses Vendégszereplések Magánélete Viszályok Mióta a Guns N’ Roses-t megalapították, 1985-ben, Rose sok zenésszel, ismert emberrel, és zenekarral viszálykodott már. Ilyen például Vince Neil, a Mötley Crüe frontembere, Kurt Cobain, Scott Weiland, korábbi barátja és zenekartársa Slash, James Hetfield, a divattervező Tommy Hilfiger, Jon Bon Jovi, az Eagles of Death Metal és Dexter Holland. Kapcsolatai A ’80-as évek közepén és a '90-es évek elején Rose viharos kapcsolatba került Erin Everlyvel, Don Everly (Everly Brothers) énekes lányával. Rose a „Sweet Child o’ Mine” című dalt Erinnek írta, a lány a videóban is megjelent. Axl és Erin 1990 április 28-án Las Vegas-ban házasodtak össze, bár kényszer alatt, miután Rose megjelent Erin házánál, azt állítva, hogy fegyver van az autójában és öngyilkos lesz, ha Erin nem megy hozzá feleségül. Rose azt akarta, hogy Dan McCafferty (a Nazareth énekese) énekeljen az esküvőn. Egy 2004-es interjúban McCafferty úgy nyilatkozott, hogy "Röhögőgörcsöt kaptam, amikor Axl arra kért, hogy énekeljem el nekik a „Love Hurts” című dalt az esküvőn, mivel tudtam, hogy a dal maga hosszabb, mint a házasságuk lesz. A Guns N’ Roses menedzsmenttől 18 ember hívott fel, hogy tegyem meg, de végül elfoglaltságra hivatkozva visszautasítottam. Valószínűleg tényleg elfoglalt lehettem.” A pár kevesebb, mint egy hónapigle nom de Tori Amos (1994) Guns N' Roses Use Your Illusion I (1991) Use Your Illusion II (1991) The Spaghetti Incident? (1993) Live Era: '87-'93 (1999) Greatest Hits (2004) Velvet Revolver Contraband (2004) Libertad (2007) (EP) (2007) The Cult Beyond Good And Evil (2001) Slash's Snakepit It's Five O'Clock Somewhere (1995) Neurotic Outsiders Neurotic Outsiders (1996) Ainsi que d'autres participations avec des groupes tels que Hawk, Jeff Paris, Wired Up, Y Kant Tori Read et Johnny Crash, et un album solo en 2003 intitulé "Hollywood Zen". Bibliographie Mick Wall, Guns N’ Roses: The Most Dangerous Band In The World, Sidgwick & Jackson, 1991. Danny Sugerman, Appetite for Destruction : the days of Guns N'Roses, 1991. Robert John, Guns N'Roses : the photographic history, 1993. Photos et vidéos inédites de Guns N'Roses sur le site du photographe Robert John Slash, Slash, autobiographie en collaboration avec Anthony Bozza, Harper Entertainment, 2007. Nick Kent, L'Envers du rock, éditions Austral, 1996. Christian Lamet, Guns N' Roses 1985-1995, éditions La Sirène, 1995. Notes et références Liens externes Guns N' Roses Naissance en novembre 1960 Batteur de hard rock Batteur américain de rock Naissance à Mission Viejo Membre de Guns N' Roses Membre de Velvet Revolveron oral promises, within the two-year statute of limitations that began with the game's October 2007 release. In March 2013, Rose was sued by an audience member who got hit by a microphone Rose threw into the crowd at the end of a show. Rose threatened legal action in 2014 over the release of Rapidfire (his pre-Hollywood Rose band) recordings, keeping them from digital storefronts for a period of time. In August 2016, former Guns N' Roses keyboardist Chris Pitman sued Rose for $125,000 in unpaid wages. The two parties settled in November 2016. Discography with Guns N' Roses Appetite for Destruction (1987) G N' R Lies (1988) Use Your Illusion I (1991) Use Your Illusion II (1991) "The Spaghetti Incident?" (1993) Chinese Democracy (2008) with Hollywood Rose The Roots of Guns N' Roses (2004) with Rapidfire Ready to Rumble EP (2014) Guest appearances The Decline of Western Civilization Part II: The Metal Years – Original Motion Picture Soundtrack by various artists (1988; "Under My Wheels" ft. Alice Cooper, Slash and Izzy Stradlin) The End of the Innocence by Don Henley (1989; "I Will Not Go Quietly") Fire and Gasoline by Steve Jones (1989; "I Did U No Wrong") Pawnshop Guitars by Gilby Clarke (1994; "Dead Flowers") Anxious Disease by The Outpatience (1996; "Anxious Disease" ft. Slash) Angel Down by Sebastian Bach (2007; "Back in the Saddle," "(Love Is) a Bitchslap," "Stuck Inside") New Looney Tunes (2018, "Rock the Rock") Filmography References External links 1962 births 20th-century American male singers 20th-century American singers 21st-century American male
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Were Scott Derrickson and Ed Wood of the same nationality?
-A: yes
-
-Q: What government position was held by the woman who portrayed Corliss Archer in the film Kiss and Tell?
-A: Chief of Protocol
-
-Q: What science fantasy young adult series, told in first person, has a set of companion books narrating the stories of enslaved worlds and alien species?
-A: Animorphs
-
-Q: Are the Laleli Mosque and Esma Sultan Mansion located in the same neighborhood?
-A: no
-
-Q: The director of the romantic comedy "Big Stone Gap" is based in what New York city?
-A: Greenwich Village, New York City
-
-Q: What year did Guns N Roses perform a promo for a movie starring Arnold Schwarzenegger as a former New York Police detective?
-A: 
-==================================================
-
-==================================================
-PROMPT #14:
-==================================================
-Random House Tower, de asemenea cunoscut ca Park Imperial Apartments, este un zgârie-nori de 52 de etaje din New York City, sediu al Random House și complex rezidențial de lux. Note Clădiri și structuri în New Yorkof up to . The residential floors are numbered 48-70 for marketing purposes. Among the first tenants were rapper P. Diddy and New York Yankees pitcher Randy Johnson. The building has two fluid tuned mass dampers, which are designed to damp building sway. Located on the 50th floor mechanical room, they have concrete walls, and measure wide by long and tall. One tank runs west-east and the other runs north-south. History When the lease of its longtime headquarters at 201 East 50th Street was ending, Random House decided to expand its headquarters and move to a newly built tower. They originally planned in 1998 to build a tower at 45th and Broadway across from its parent company Bertelsmann's headquarters at 1540 Broadway with a neon-lighted skyway across 45th Street connecting them, but after long negotiations the owner of the property withdrew from the plans. Looking for an alternative, the company spoke with various developers, who were already planning apartment buildings, and in 1999 decided for the project of Stephen M. Ross at Broadway and 55th Street, which was already under construction. Subsequently, they assigned architects to develop a new design. While the building has functioned as Random House's headquarters ever since, the 2013 newly formed Penguin Random House parent company initially worked from SoHo at 345 and 375 Hudson Street. In 2016, two years before the lease of the Random House Tower would end, the company extended the contract till 2033 and also moved to the tower of its subsidiary. There, Penguin Random House occupies andThe Random House Tower, also known as the Park Imperial Apartments, is a 52-story mixed-use tower in Manhattan, New York City. It is owned by real estate companies SL Green Realty and Ivanhoé Cambridge, with the office portion leased as the headquarters to book publisher Random House from the beginning and its parent company Penguin Random House (formed in 2013) since 2016, giving the building its name; a luxury apartment complex follows above the offices. The PRH entrance is on Broadway and goes up to 27 floors, while the apartment complex entrance is on West 56th Street. Rising to 684 ft (208 m), it is the 77th tallest building in New York City. Description The complex is on a trapezoidal block between 55th Street and 56th Street and follows the angle of Broadway. It has jagged setbacks to improve the views of Central Park. The Random House Tower is divided into two parts: the office floors on the 2nd through 26th stories, and the residential floors between the 27th and 51st stories. The first floor contains of retail space. Separate architects designed each of the sections. Skidmore, Owings & Merrill designed the office portion, which has a steel frame. Ismael Leyva Architects and Adam D. Tihany designed the residential portion, which has a concrete frame. The two sections do not entirely line up, and trusses were built on the 26th and 27th floor to transfer the load. The residential section of the tower has 130 apartments. The apartments have ten-foot ceilings, and there are five penthouses
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Were Scott Derrickson and Ed Wood of the same nationality?
-A: yes
-
-Q: What government position was held by the woman who portrayed Corliss Archer in the film Kiss and Tell?
-A: Chief of Protocol
-
-Q: What science fantasy young adult series, told in first person, has a set of companion books narrating the stories of enslaved worlds and alien species?
-A: Animorphs
-
-Q: Are the Laleli Mosque and Esma Sultan Mansion located in the same neighborhood?
-A: no
-
-Q: The director of the romantic comedy "Big Stone Gap" is based in what New York city?
-A: Greenwich Village, New York City
-
-Q: Are Random House Tower and 888 7th Avenue both used for real estate?
-A: 
-==================================================
-
-==================================================
-PROMPT #15:
-==================================================
-of Beckham through suspension and Sol Campbell and Steven Gerrard through injury, in a 2–1 win against Poland. In January 2006, Eriksson was recorded saying he would be willing to leave England to manage Aston Villa if England won the World Cup, after being duped into believing that a wealthy Arab would buy the club, and wanted him as manager. The wealthy "Arab" was in fact the "Fake Sheikh" Mazher Mahmood, an undercover News of the World reporter. On 23 January, The FA announced that Eriksson would leave his job after the 2006 World Cup, and it was thought that the News of the World allegations played a part in this decision. This was later denied by both parties, with Eriksson explaining that there was a prior arrangement to terminate his contract immediately after the World Cup. In March 2006, The Guardian reported that South Africa became the first potential employer publicly to register its interest in Eriksson. That job, however, went to Carlos Alberto Parreira. The tournament England finished top of Group B at the finals, beating Paraguay and Trinidad and Tobago, before a 2–2 draw with Sweden, although the English press considered their performances far from satisfactory. In the second round, a David Beckham trademark free kick was enough to see Eriksson's England past Ecuador in a lacklustre 1–0 encounter affected by very high temperatures. Eriksson, however, once again fell to nemesis Luiz Felipe Scolari's Portugal. They defeated England 3–1 on penalties with the score 0–0 after extra time, with Beckham lost to injuryprevious season as Ferguson's assistant at Manchester United. In late July 2003, the club went on a tour of the Far East as part of pre-season training, but also to cash in on Beckham's huge marketing appeal in Asia, where he enjoyed tremendous following. Comparing his reception upon arriving at Kunming Airport in south China to Beatlemania, Marca ran the headline, "Beckham-mania in China". After the opening game in Beijing the tour featured games in Hong Kong, Tokyo and Bangkok. Real's brand recognition in that part of the globe was already well established as the club made financially successful trips to Asia during previous off-seasons. The presence of a global marketing icon such as Beckham, however, made this particular tour a financial smash for los Merengues. Shortly after his transfer to Real, Beckham also ended his relationship with agent Tony Stephens of SFX Europe, who had guided him through his career until that point, including helping to engineer Beckham's move from Manchester to Madrid. Beckham signed on with Simon Fuller and his company 19 Entertainment, which already managed the career of Victoria. Beckham also appointed close friend Terry Byrne to be his personal manager. In late August 2003, Real Madrid won the Spanish Super Cup over two legs versus RCD Mallorca, with Beckham scoring the final goal in a 3–0 return leg win at home, thus setting the stage for the start of the league season. Playing in a star-laden team which included three former FIFA World Player of the Year recipients, Zinedine Zidane, Ronaldo andJim Blyth played scintillating attacking football that swept many teams aside, often by large margins of victory, although they failed to win any silverware or qualify for the UEFA Cup. He later played for Fulham, AFC Bournemouth and finally joined Cambridge United in July 1986. Managerial career Beck's career was ended by injury in 1989, and he was appointed assistant manager at Cambridge. Following the resignation of Chris Turner he was appointed manager January 1990, with Cambridge mid table in the Fourth Division. In his first season United won the Fourth Division playoffs, and reached the quarter-finals of the FA Cup – a rare achievement for a Fourth Division side. The following season (1990–91) they won the Third Division title after an excellent finish to the season saw them climb from 10th to top in less than three months, and they repeated the feat of reaching the FA Cup quarter-finals. During the Summer, Beck turned down the opportunity to manage Leicester City, saying he wanted to take the club all the way to the First Division. In fact he very nearly achieved this, as in 1991–92 they finished 5th in the Second Division, the highest position ever achieved by the club, but lost to Leicester City in the play-off semi-finals, ending their hopes of a third successive promotion (something which has still yet to be achieved in the English Football League) and denying them the chance of becoming Premier League founder members. Beck's long ball playing style, despite its success, was heavily criticised in the media,
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Were Scott Derrickson and Ed Wood of the same nationality?
-A: yes
-
-Q: What government position was held by the woman who portrayed Corliss Archer in the film Kiss and Tell?
-A: Chief of Protocol
-
-Q: What science fantasy young adult series, told in first person, has a set of companion books narrating the stories of enslaved worlds and alien species?
-A: Animorphs
-
-Q: Are the Laleli Mosque and Esma Sultan Mansion located in the same neighborhood?
-A: no
-
-Q: The director of the romantic comedy "Big Stone Gap" is based in what New York city?
-A: Greenwich Village, New York City
-
-Q: The football manager who recruited David Beckham managed Manchester United during what timeframe?
-A: 
-==================================================
-
-==================================================
-PROMPT #16:
-==================================================
-Lake Refuge Luedtke State Wildlife Management Area Perch Creek State Wildlife Management Area State Wildlife Management Area Truman Wildlife Area Lakes Amber Lake Big Twin Lake Bright Lake Budd Lake Buffalo Lake (Cedar Township) Buffalo Lake (Rutland Township) Canright Lake Cedar Lake Clam Lake Clayton Lake Clear Lake Creek Lake Eagle Lake East Chain Lake Fish Lake (Cedar Township) Fish Lake (Lake Belt Township) Fox Lake Hall Lake High Lake Iowa Lake (part) Kiester Lake Lake Charlotte Lake George Lake Imogene Lake Seymour Little Hat Lake Little Twin Lake Martin Lake Mud Lake Murphy Lake North Lake North Silver Lake Okamanpeedan Lake (part) Perch Lake Pierce Lake Rose Lake Round Lake (Fox Lake Township) Round Lake (Waverly Township) Sager Lakes Sisseton Lake South Silver Lake Susan Lake Temperance Lake Tuttle Lake Watkins Lake Wilmert Lake Demographics 2000 census As of the 2000 census, there were 21,802 people, 9,067 households, and 6,047 families in the county. The population density was 30.6/sqmi (11.8/km2). There were 9,800 housing units at an average density of 13.8/sqmi (5.31/km2). The racial makeup of the county was 97.22% White, 0.25% Black or African American, 0.10% Native American, 0.42% Asian, 0.02% Pacific Islander, 1.26% from other races, and 0.73% from two or more races. 1.93% of the population were Hispanic or Latino of any race. 51.6% were of German, 11.9% Norwegian, 5.9% Swedish and 5.0% English ancestry. There were 9,067 households, out of which 29.80% had children under the age of 18 living with them, 56.60% were married couples living together, 7.20% had athe table for some people during a period when the local economy was in a downward slide. In 1969, the Texas Utilities Generating Company located a new power plant near Fairfield called Big Brown Power Plant. A dam was built to create Fairfield Lake to provide stored water for a cooling system for the plant. Fairfield Lake State Park was established around the lake and opened to the public in 1976. Big Brown was shut down in February 2018. Geography According to the U.S. Census Bureau, the county has a total area of , of which are land and (1.6%) are covered by water. Major highways Interstate 45 U.S. Highway 79 U.S. Highway 84 U.S. Highway 287 State Highway 14 State Highway 75 State Highway 164 State Highway 179 Adjacent counties Henderson County (north) Anderson County (northeast) Leon County (southeast) Limestone County (southwest) Navarro County (northwest) Demographics Note: the US Census treats Hispanic/Latino as an ethnic category. This table excludes Latinos from the racial categories and assigns them to a separate category. Hispanics/Latinos can be of any race. As of the census of 2010, 19,816 people, 6,588 households, and 4,664 families were residing in the county. The population density was 20 people/sq mi (8/km2). The 8,138 housing units averaged 9/sq mi (4/km2). The racial makeup of the county was 73.1% White, 16.1% African American, 0.7% Native American, 0.3% Asian, 8.1% from other races, and 1.6% from two or more races. About 13.6% of the population were Hispanic or Latino of any race. Of the 7,259 households,Brown State Fishing Lake (sometimes also known as Brown State Fishing Lake And Wildlife Area) is a protected area in Brown County, Kansas in the United States. The lake is 62 acres (0.25 km2) in area and up to 13 feet (4 m) deep. The area was formerly known as Brown County State Park, and is 8 miles (13 km) east of Hiawatha, Kansas. References Protected areas of Brown County, Kansas Kansas state fishing lakes Bodies of water of Brown County, Kansas Reservoirs in Kansas
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Were Scott Derrickson and Ed Wood of the same nationality?
-A: yes
-
-Q: What government position was held by the woman who portrayed Corliss Archer in the film Kiss and Tell?
-A: Chief of Protocol
-
-Q: What science fantasy young adult series, told in first person, has a set of companion books narrating the stories of enslaved worlds and alien species?
-A: Animorphs
-
-Q: Are the Laleli Mosque and Esma Sultan Mansion located in the same neighborhood?
-A: no
-
-Q: The director of the romantic comedy "Big Stone Gap" is based in what New York city?
-A: Greenwich Village, New York City
-
-Q: Brown State Fishing Lake is in a country that has a population of how many inhabitants ?
-A: 
-==================================================
-
-==================================================
-PROMPT #17:
-==================================================
-The Vermont Catamounts are the varsity intercollegiate athletic programs of the University of Vermont, based in Burlington, Vermont, United States. The school sponsors 18 athletic programs (8 men's, 10 women's), most of which compete in the NCAA Division I America East Conference (AEC), of which the school has been a member since 1979. The men's and women's ice hockey programs compete in Hockey East. The men's and women's alpine and nordic skiing teams compete in the Eastern Intercollegiate Ski Association (EISA). The school's athletic director is Jeff Schulman. The Catamounts have won six national championships, all in skiing. The program's mascot is Rally and colors are green and gold. Teams Vermont is one of only four NCAA Division I schools that do not sponsor volleyball or baseball. The other three are Boston University, Detroit Mercy, and Drexel. Basketball The men's basketball team won the 2003, 2004, 2005, 2010, 2012, 2017, 2019, 2022, and 2023, America East Conference Championships, qualifying for the NCAA Tournament in each of those seasons. In the 2005 NCAA tournament, the team, seeded #13 in the Austin Regional, upset #4 Syracuse 60–57 in overtime. Sports Illustrated named the upset one of the top 10 NCAA Tournament upsets of the 2000s. Ice hockey The men's ice hockey team, coached by Kevin Sneddon, appeared in the 2009 Men's Frozen Four. It has also qualified for the NCAA Tournament in 1996, 1997, 2010, and 2014. UVM's men's hockey team has produced 18 NHL players in its history. UVM alumni currently in the NHL include Viktor StålbergThe Vermont Catamounts football program was the intercollegiate American football team for the University of Vermont located in Burlington, Vermont. The team competed in the NCAA Division I and were members of the Yankee Conference. The school's first football team was fielded in 1886. The football program was discontinued at the conclusion of the 1974 season. Vermont fields a team at the club football level, in a conference that also uses the Yankee Conference name. History The first intercollegiate game in the state of Vermont happened on November 6, 1886, between Dartmouth and Vermont in Burlington. Dartmouth won 91 to 0. Vermont reached a level of success after coach "Dud" Drake in the 1907 and 1908 seasons. The 1907 team fought Dartmouth to a scoreless tie, and the 1908 team gave Cornell a scare. In 2007, The University of Vermont brought football back as a club sport under the Student Government Association by Doug Deluca. The University of Vermont Club football team is currently playing in the NCFA and is led by head coach Jeff Porter, defensive coordinator Jason Paul and offensive coordinator Jack Leclerc. As of recent, the team has sustained some success under head coach Jeff Porter. During the 2019 NCFA season, the Catamounts went undefeated in the North Atlantic conference until losing in their conference championship game to Sacred Heart University. The Catamounts are currently ranked fifth in the NCFA. All-time Yankee Conference records This table reflects the results of Yankee Conference matchups when both Vermont and its opponent were members of theThe Vermont Catamounts men's soccer team represents the University of Vermont in all NCAA Division I men's college soccer competitions. The team competes in the America East Conference. The team currently practices at Virtue Field adjacent to Centennial Field on campus. Roster Year by year results Note: Vermont competed in the Yankee Conference from 1964 to 1979 and has competed in the America East Conference since 1988. {| class="wikitable" style="text-align:center;" |- align="center" See also Vermont Catamounts Vermont Catamounts women's soccer College soccer References External links Soccer clubs in Vermont 1964 establishments in Vermont Association football clubs established in 1964
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Were Scott Derrickson and Ed Wood of the same nationality?
-A: yes
-
-Q: What government position was held by the woman who portrayed Corliss Archer in the film Kiss and Tell?
-A: Chief of Protocol
-
-Q: What science fantasy young adult series, told in first person, has a set of companion books narrating the stories of enslaved worlds and alien species?
-A: Animorphs
-
-Q: Are the Laleli Mosque and Esma Sultan Mansion located in the same neighborhood?
-A: no
-
-Q: The director of the romantic comedy "Big Stone Gap" is based in what New York city?
-A: Greenwich Village, New York City
-
-Q: The Vermont Catamounts men's soccer team currently competes in a conference that was formerly known as what from 1988 to 1996?
-A: 
-==================================================
-
-==================================================
-PROMPT #18:
-==================================================
-Opéra orchestra. In 1828, aged 17, Ambroise joined him in Paris, where he was admitted as a student by the Conservatoire de Paris. He studied the piano with Pierre Zimmerman and harmony and counterpoint with Victor Dourlen. He won premiers prix in these subjects in 1829 and 1830. He went on to study the piano with Friedrich Kalkbrenner, and composition with Jean-François Lesueur and Auguste Barbereau. In 1832, at his second attempt, Thomas won France's premier music prize, the Grand Prix de Rome, with his cantata Hermann et Ketty. The prize brought him three years' study at the Villa Medici, the French Academy in Rome. During his time there he became friendly with the painter Ingres, the head of the academy, with whom he shared an admiration for both Mozart and Beethoven; he also met Berlioz, who encouraged him and wrote about him favourably. During his Italian sojourn he wrote chamber music – a piano trio, a string quintet and a string quartet – and a set of six songs, Souvenirs d'Italie. After leaving Rome, Thomas stayed briefly in Germany, before returning to Paris in 1835, when he began writing for the stage. Composing career The first opera Thomas composed was La double échelle (The Double Ladder, 1837), a one-act comedy, praised by Berlioz for its "extreme vivacity and wit". It was produced at the Opéra-Comique, received 247 performances, and in the next few years was given in Brussels, New Orleans, Berlin, Vienna and London. His first full-length opera, Le perruquier de la Régence (The RegencyCharles Louis Ambroise Thomas (; 5 August 1811 – 12 February 1896) was a French composer and teacher, best known for his operas Mignon (1866) and Hamlet (1868). Born into a musical family, Thomas was a student at the Conservatoire de Paris, winning France's top music prize, the Prix de Rome. He pursued a career as a composer of operas, completing his first opera, La double échelle, in 1837. He wrote twenty further operas over the next decades, mostly comic, but he also treated more serious subjects, finding considerable success with audiences in France and abroad. Thomas was appointed as a professor at the Conservatoire in 1856, and in 1871 he succeeded Daniel Auber as director. Between then and his death at his home in Paris twenty-five years later, he modernised the Conservatoire's organisation while imposing a rigidly conservative curriculum, hostile to modern music, and attempting to prevent composers such as César Franck and Gabriel Fauré from influencing the students of the Conservatoire. Thomas' operas were generally neglected during most of the 20th century, but in more recent decades they have experienced something of a revival both in Europe and the US. Life and career Early years Thomas was born in Metz, the youngest of four children of Martin Thomas (1770–1823) and his wife, Jeanne, née Willaume (1780–1866), both music teachers. By the age of ten he was already an experienced pianist and violinist. When he was twelve his father died, and Ambroise's elder brother Charles moved to Paris, where he played the cello in theThis is a list of the complete operas of the French opera composer Ambroise Thomas (1811–1896). All premieres took place in Paris unless otherwise noted. References Forbes, Elizabeth (1992), 'Thomas, Ambroise' in The New Grove Dictionary of Opera, ed. Stanley Sadie (London) Wild, Nicole; Charlton, David (2005). Théâtre de l'Opéra-Comique Paris: répertoire 1762-1972. Sprimont, Belgium: Editions Mardaga. . Lists of operas by composer Lists of compositions by composer
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Were Scott Derrickson and Ed Wood of the same nationality?
-A: yes
-
-Q: What government position was held by the woman who portrayed Corliss Archer in the film Kiss and Tell?
-A: Chief of Protocol
-
-Q: What science fantasy young adult series, told in first person, has a set of companion books narrating the stories of enslaved worlds and alien species?
-A: Animorphs
-
-Q: Are the Laleli Mosque and Esma Sultan Mansion located in the same neighborhood?
-A: no
-
-Q: The director of the romantic comedy "Big Stone Gap" is based in what New York city?
-A: Greenwich Village, New York City
-
-Q: Are Giuseppe Verdi and Ambroise Thomas both Opera composers ?
-A: 
-==================================================
-
-==================================================
-PROMPT #19:
-==================================================
-in the Carter Administration. He was appointed Deputy Assistant Secretary for Health in the U.S. Department of Health, Education and Welfare (now Department of Health and Human Services), and served through four Administrations—Presidents Jimmy Carter, Ronald Reagan, George H. W. Bush, and William J. Clinton—while also serving as Assistant Surgeon General and Director of the U.S. Office of Disease Prevention. During this period, McGinnis chaired the Secretary's Task Force on Smoking and Health, which resulted in Secretary Califano's 1974 decision to elevate the federal leadership profile on tobacco control; led the development of Healthy People, the nation's prevention agenda; developed, with USDA, the first Dietary Guidelines for Americans; and created the U.S. Preventive Services Task Force in 1984 to develop an ongoing systematic and comprehensive assessment of the effectiveness of clinical preventive services, thereby helping to pioneer the broader advance of evidence-based medicine. Healthy People is perhaps the most widely known of McGinnis’ policy contributions. Begun when he was serving as Chair of the Secretary's Task Force on Disease Prevention and Health Promotion for HEW Secretary Joseph Califano and Surgeon General Julius Richmond in 1978, it started as an effort to develop a Surgeon General's report to serve as a U.S. answer to A New Perspective on the Health of Canadians, a groundbreaking report released in 1974 by Canadian Minister of Health Marc Lalonde. McGinnis felt the U.S. report could go beyond simply marshaling the scientific evidence underpinning the power of prevention, to establish quantifiable, evidence-based goals for improving the health of Americans through preventive interventions.to the rank of colonel and was awarded the Bronze Star Medal and the Legion of Merit. At the end of the war, Egeberg treated the Prime Minister of Japan, Hideki Tojo who had shot himself. Tojo was later hanged. Academic positions After the war, Egeberg was Chief of Medicine at Wadsworth General Hospital, head of County-University of Southern California Medical Center (1959) and Dean, USC School of Medicine (1964). After working for Nixon, Egeberg took a professorial chair at Georgetown University. In his later career, he advocated for the advanced training of doctors in geriatric medicine. Nixon administration From 1969 to 1971, Egeberg was the Assistant Secretary for health and scientific affairs in the Department of Health, Education and Welfare. In this role, he was an advocate for expanding public health care and argued for the financing of medical programs (through the Office of Health Care Financing). During this time, Egeberg visited the USSR and developed a professional relationship with Boris Vasilevich Petrovsky, the Soviet minister of health. In the spring of 1970, while requesting $100 million to train more doctors and nurses, Egeberg alleged the Health, Education and Welfare Secretary, Robert H. Finch, and the department's Under Secretary, John C. Veneman, were not considering his advice on health matters and that Nixon's advisers were ignoring him. He said, "The White House just doesn't appreciate, doesn't know, what is going on in the health field." also, "I just can't get through to Ehrlichman," (referring to John D. Ehrlichman, the president's chief aide for domestic matters.)Roger Olaf Egeberg, M.D. (13 November 1902 – 13 September 1997 Washington, D.C.) was an American medical educator, administrator and advocate of public health. He was General Douglas MacArthur's personal physician during World War II in the Pacific theater. His other roles included Assistant Secretary for Health and Scientific Affairs in the Department of Health, Education, and Welfare (now the United States Department of Health and Human Services) during the Nixon administration and Dean of the University of Southern California's medical school. Early life and family Egeberg was born in Chicago, Illinois to Hans Olaf Egeberg (1875-1932) and Ulrikka Rostrup Egeberg (1875-1932) (née Nielsen), a Norwegian immigrant family. Egeberg married a Canadian sculptor, Margaret Chahoon (25 August 1904 - 31 March 2011), in 1929. Together, they had a son and three daughters. Education and career Egeberg received a bachelor's degree in 1925 from Cornell University where he was a member of the Quill and Dagger society. He earned his medical degree in 1929 at Northwestern University school of medicine. He became an internist and joined the faculty of Western Reserve University School of Medicine. Military service During World War II, Egeberg was a member of the 4th General Hospital which was sent to Australia. From there he transferred to Milne Bay, Papua New Guinea (PNG) where he organised field stations and was a malaria control officer. Through determined attempts to control illnesses such as malaria and sexually transmitted diseases in PNG, Egeberg was noticed by MacArthur who made him his personal physician and aiLine truncated
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Were Scott Derrickson and Ed Wood of the same nationality?
-A: yes
-
-Q: What government position was held by the woman who portrayed Corliss Archer in the film Kiss and Tell?
-A: Chief of Protocol
-
-Q: What science fantasy young adult series, told in first person, has a set of companion books narrating the stories of enslaved worlds and alien species?
-A: Animorphs
-
-Q: Are the Laleli Mosque and Esma Sultan Mansion located in the same neighborhood?
-A: no
-
-Q: The director of the romantic comedy "Big Stone Gap" is based in what New York city?
-A: Greenwich Village, New York City
-
-Q: Roger O. Egeberg was Assistant Secretary for Health and Scientific Affairs during the administration of a president that served during what years?
-A: 
-==================================================
-
-==================================================
-PROMPT #20:
-==================================================
-Robert Cesar Childers (sprich tschillders; * 1838 in Nizza; † 28. Juli 1876 in London) gilt als einer der herausragenden Kenner des Pali und des Buddhismus in seiner Zeit. Er war der Vater des Schriftstellers und Anhängers der irischen Unabhängigkeitsbewegung Robert Erskine Childers und der Großvater des irischen Präsidenten Erskine Hamilton Childers. Childers studierte in Oxford und ging 1860 nach Ceylon. Während eines mehrjährigen Aufenthalts in Ceylon als englischer Zivilbeamter machte er sich mit Hilfe eines Singhalesen mit dem Pâli bekannt. 1864 kehrte er aus gesundheitlichen Gründen nach England zurück, wo er 1872 Unterbibliothekar an der Bibliothek des India Office (Indian Office Library) und später Professor des Pâli und der buddhistischen Literatur am University College in London wurde. Im Journal der Asiatic Society gab er mehrere Pâlitexte mit Übersetzungen sowie Untersuchungen über das Singhalesische, die einheimische Sprache von Ceylon, heraus, die er als Tochter des Sanskrits nachzuweisen versuchte. Sein Hauptwerk ist das preisgekrönte »Dictionary of the Pali language« (London 1875), das eine neue Epoche im Studium der Pâliliteratur und des Buddhismus begründete. An der Herausgabe einer im Manuskript fertigen Pâligrammatik wurde er durch den Tod gehindert. Childers starb am 28. Juli 1876 in London. Quellen Brockhaus Enzyklopädie. 14. A., Bd. 4. 1908. Indologe Buddhologe Brite Engländer Geboren 1838 Gestorben 1876 MannRobert Caesar Childers (12 February 1838 – 25 July 1876) was a British Orientalist scholar, compiler of the first Pāli-English dictionary. Childers was the husband of Anna Barton of Ireland. He was the father of Irish nationalist Erskine Childers and grandfather to the fourth President of Ireland, Erskine Hamilton Childers. Early life and marriage He was born in Nice, Piedmont-Sardinia, the son of Canon Charles Childers, chaplain to the English colony in Nice. His grandmother was Selina Eardley. In 1857, at the age of nineteen, he began the study of Hebrew at Wadham College, Oxford, graduating in 1861. After some travels in Ireland, he soon married Anna Barton of Glendalough House. The Bartons of Wicklow, Ireland, were a very wealthy and respected family in Irish farming and politics. Pali dictionary He then moved to Ceylon for an official position in the civil service. During this period he studied Sinhalese culture, particularly the Pali language. In 1869 he published the first Pali text in Britain, and began to work on a Pali dictionary, which was published 1872–75. In 1872 he was appointed sub-librarian at the India Office, and in the following year he became the first professor of Pali and Buddhist literature at University College, London. In 1876 Childers' dictionary was awarded the Volney Prize by the Institute of France. He died at the age of 38 in London the same year, apparently having succumbed to tuberculosis which had been troubling him for many years, and which grew worse after Childers contracted a cold in the earlyRobert Erskine Childers DSC (25 June 1870 – 24 November 1922), usually known as Erskine Childers (), was an English-born Irish writer, politician, and militant. His works included the influential novel The Riddle of the Sands. Starting as an ardent Unionist, he later became a supporter of Irish Republicanism and smuggled guns into Ireland in his sailing yacht Asgard. He was executed by the authorities of the nascent Irish Free State during the Irish Civil War. He was the son of British Orientalist scholar Robert Caesar Childers; the cousin of Hugh Childers and Robert Barton; and the father of the fourth President of Ireland, Erskine Hamilton Childers. Early life Childers was born in Mayfair, London, in 1870. He was the second son of Robert Caesar Childers, a translator and oriental scholar from an ecclesiastical family, and Anna Mary Henrietta Barton, from an Anglo-Irish landowning family of Glendalough House, Annamoe, County Wicklow, with interests in France such as the winery that bears their name. When Erskine was six, his father died from tuberculosis and, although seemingly healthy, Anna was confined to an isolation hospital, where she died six years later. The five children were sent to the Bartons, the family of their mother's uncle, at Glendalough, County Wicklow. They were treated kindly there and Erskine grew up knowing and loving Ireland, albeit at that stage from the comfortable viewpoint of the "Protestant Ascendancy". According to his biographer Michael Hopkinson, it was the personal tension caused by his innate belief in English superiority, in conflict with this new
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Were Scott Derrickson and Ed Wood of the same nationality?
-A: yes
-
-Q: What government position was held by the woman who portrayed Corliss Archer in the film Kiss and Tell?
-A: Chief of Protocol
-
-Q: What science fantasy young adult series, told in first person, has a set of companion books narrating the stories of enslaved worlds and alien species?
-A: Animorphs
-
-Q: Are the Laleli Mosque and Esma Sultan Mansion located in the same neighborhood?
-A: no
-
-Q: The director of the romantic comedy "Big Stone Gap" is based in what New York city?
-A: Greenwich Village, New York City
-
-Q: Which writer was from England, Henry Roth or Robert Erskine Childers?
-A: 
-==================================================
-

benchmarks/data/prompts_g5/prompt_dump_nq_hnsw.txt

@@ -1,484 +0,0 @@
-=== Prompt Dump for NQ + HNSW ===
-Total prompts: 50
-Showing first 20 prompts:
-
-==================================================
-PROMPT #1:
-==================================================
-First Time<span style="font-size=50%>...</span>" exploded in popularity following the November 1971 release of Play Misty For Me. This persuaded Atlantic Records to issue the track as a single - trimmed by a minute - in February 1972: the track became a smash hit single in the United States, reaching No. 1 for six weeks on both the Billboard Hot 100 and easy listening charts in the spring of 1972, with a No. 4 R&B chart peak. Reaching No. 14 on the UK Singles Chart, Flack's "The First Time<span style="font-size=50%>...</span>" was No. 1 for three weeks on the singles chart in Canada's RPM magazine. "The First Time Ever I Saw Your Face" was played as the wake-up music on flight day 9 to the astronauts aboard Apollo 17 on their last day in Lunar orbit (Friday, December 15, 1972) before returning to Earth, thus ending the last human explorations of the Moon. The use of the song was most likely a reference to the "face" of the Moon below the spacecraft. Chart history Weekly charts Roberta Flack Year-end charts All-time charts Celine Dion Leona Lewis Matt Cardle See also List of Billboard Hot 100 number-one singles of 1972 List of number-one adult contemporary singles of 1972 (U.S.) List of number-one singles in Australia during the 1970s List of number-one singles of 1972 (Canada) References External links Superseventies.com - with quotes from Roberta Flack and information on the song's background 1957 songs 1972 singles 2000 singles Songs written by Ewan MacColl Roberta Flack songs Celine Dion songs The Kingstonthe Apollo 11 mission to the Moon. The song had been recorded three weeks earlier. The Sprague Electric Company delivered a diameter silicon disc to NASA, containing 73 "messages of good will from the leaders of the world's nations to be flown and left on the Moon," that had been inscribed microscopically. The medallion was placed on a package adhered to one of the sleeves of Buzz Aldrin's spacesuit, and was almost forgotten until Aldrin was climbing back into the lunar module before for departure on July 21. Neil Armstrong reminded Aldrin of "that package" and Aldrin tossed it on to the lunar surface. Born: David Tao (Tao Xuzhong), Taiwanese R&B singer and songwriter; in Hong Kong July 12, 1969 (Saturday) Nearly four months after its cancellation by the American NBC TV network, Star Trek was introduced to British television viewers. Its run on BBC-1 began with the pilot episode "Where No Man Has Gone Before" (unlike the original US run, which had opted for "The Man Trap"). Filling in as a summer replacement for Doctor Who between the Doctor's sixth and seventh seasons, the programme was shown in the Doctor Who 5:15 p.m. time slot between afternoon sports and the 10-minute BBC news and weather update. July 13, 1969 (Sunday) The Soviet Union launched the unmanned lunar spacecraft Luna 15, three days before the scheduled liftoff of the American Apollo 11 manned mission to the Moon, with the objective of performing a sample-return mission and bringing back the first lunar soil (a "Moon rock") aheadto 19 December 1972. The 24 people who have flown to the Moon are the only people who have traveled beyond low Earth orbit. Ten of them are alive. Jim Lovell, John Young, and Eugene Cernan are the only three people to have flown to the Moon twice. Young and Cernan each set foot on it during their respective second lunar missions, while Lovell is the only person to have flown to the Moon twice without landing. During Cernan's first lunar mission on Apollo 10, he tied the present record set by Bill Anders on Apollo 8 as the youngest person to fly to the Moon. Each was 35 years and 65 days old on his launch date and 35 years and 68 days old when he entered lunar orbit. The oldest person to fly to the Moon was Alan Shepard, who walked on its surface during the Apollo 14 mission. Shepard was 47 years and 74 days old on his launch date and 47 years and 78 days old when he entered lunar orbit. Jim Lovell and Fred Haise were scheduled to walk on the Moon during the Apollo 13 mission, but the lunar landing was aborted following a major malfunction en route to the Moon. Haise was again scheduled to walk on the Moon as commander of Apollo 19, but Apollo 18 and Apollo 19 were canceled on September 2, 1970. Because of Apollo 13's free-return trajectory, Lovell, Swigert and Haise flew higher above the Moon's 180° meridian (opposite Earth) than anyone else has
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: where did they film hot tub time machine?
-A: Fernie Alpine Resort
-
-Q: who has the right of way in international waters?
-A: Neither vessel
-
-Q: who does annie work for attack on titan?
-A: Marley
-
-Q: when was the immigration reform and control act passed?
-A: November 6, 1986
-
-Q: when was puerto rico added to the usa?
-A: 1950
-
-Q: when was the last time anyone was on the moon?
-A:
-==================================================
-
-==================================================
-PROMPT #2:
-==================================================
-My Brother", the first use of the phrase exactly as it is rendered in the song title. In the 1940s, the words, adapted as "He ain't heavy, Father, he's my brother", were taken as a slogan for Boys Town children's home by founder Father Edward Flanagan. According to the Boys Town website, the phrase as used by Boys Town was said to Fr. Flanagan in 1918 by one of the residents while carrying another up a set of stairs. The boy being carried is said to have had polio and worn leg braces. The Hollies version The Hollies' recorded the song in June 1969 at the EMI Studios (now Abbey Road Studios), with Allan Clarke on lead vocals. Elton John, who was working as a session musician at the time, played the piano on the song, as well their next single, "I Can't Tell the Bottom from the Top". The song was released on 26 September 1969 and reached No. 3 in the UK, and No. 7 in the US. The song was re-released in August 1988 in the UK following its use in a television advertisement for Miller Lite beer. It reached the No. 1 spot in the UK chart for two weeks in September 1988. Weekly charts Year-end charts Neil Diamond version The Neil Diamond version entered at No. 68 on the Hot 100 on November 7, 1970 (UNI Records, 55264, length 4:09). The flip side was "Free Life". The song appears on Diamond's album Tap Root Manuscript, which was released in November 1970."He Ain't Heavy, He's My Brother" is a 1969 song by the band the Hollies. He Ain't Heavy, He's My Brother may also refer to: He Ain't Heavy, He's My Brother (album), a 1969 Hollies album, also known as Hollies Sing Hollies and including the song of the same name He Ain't Heavy, He's My Brother (film), a Hong Kong film, also known as He Ain't Heavy, He's My Father or 新難兄難弟 "He Ain't Heavy, He's My Brother", an episode of Rizzoli & Isles "He Ain't Heavy, He's My Brother", an episode of On Our Own (1994 TV series) The caption of a logo adopted by Boys Town in 1943 See also He Ain't Heavy (disambiguation) He's No Heavy, He's My Brother (disambiguation) He's Heavy, He's My Brother (disambiguation) He Ain't Heavy, He's My Father (disambiguation) "He Ain't a Hottie, He's My Brother", an episode of Hannah Montana "He Ain't Heavy, He's My Bother", an episode of Heathcliff (1984 TV series) "He Ain't Heavy, He's Dee's Brother", an episode of Moesha "He Ain't Heavy, He's Willie's Brother", an episode of ALF "Lee Ain't Heavy, He's My Brother", an episode of Three's Company"He Ain't Heavy, He's My Brother" is a ballad written by Bobby Scott and Bob Russell. Originally recorded by Kelly Gordon in 1969, the song became a worldwide hit for the Hollies later that year and also a hit for Neil Diamond in 1970. It has been recorded by many artists in subsequent years. The Hollies' version was re-released in 1988 and again was a major hit in the UK. Scott and Russell were introduced to each other by Johnny Mercer, at a California nightclub. Although Russell was dying of lymphoma and the pair met only three times, they managed to collaborate on the song. Title James Wells, Moderator of the United Free Church of Scotland, tells the story of a little girl carrying a big baby boy in his 1884 book The Parables of Jesus. Seeing her struggling, someone asked if she wasn't tired. With surprise she replied: "No, he's not heavy; he's my brother." In a 1918 publication by Ralph Waldo Trine titled The Higher Powers of Mind and Spirit, Trine relates the following anecdote: "Do you know that incident in connection with the little Scottish girl? She was trudging along, carrying as best she could a boy younger, but it seemed almost as big as she herself, when one remarked to her how heavy he must be for her to carry, when instantly came the reply: 'He's na heavy. He's mi brither.'" The first editor of Kiwanis magazine, Roe Fulkerson, published a column in September 1924 carrying the title "He Ain't Heavy, He's
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: where did they film hot tub time machine?
-A: Fernie Alpine Resort
-
-Q: who has the right of way in international waters?
-A: Neither vessel
-
-Q: who does annie work for attack on titan?
-A: Marley
-
-Q: when was the immigration reform and control act passed?
-A: November 6, 1986
-
-Q: when was puerto rico added to the usa?
-A: 1950
-
-Q: who wrote he ain't heavy he's my brother lyrics?
-A:
-==================================================
-
-==================================================
-PROMPT #3:
-==================================================
-the Bastards" received immense critical acclaim, with many calling it one of the best television episodes of all time. Critics cited the size and scope of the battle in the North and Daenerys's scene with her dragons at the beginning of the episode. It has a 98% rating on the review aggregator website Rotten Tomatoes from 65 reviews, with an average score of 9.08 out of 10. According to the site consensus, "'Battle of the Bastards' delivered one of the greatest battle sequences in the show's history, and some savagely satisfying vengeance as well." IGN's Matt Fowler gave the episode a 10 out of 10, writing in his review, "At this point, it seems like the pattern when it comes to the ninth episode of a given season of Game of Thrones is tragic death, amazing battle, tragic death, amazing battle, and so on. And this being Season 6, it was time to land on a phenomenal clash of swords and shields — and 'Battle of the Bastards' certainly delivered." He continued, "Game of Thrones did not disappoint when it came to this season's great northern battle, as Jon and Sansa's differences were spectacularly highlighted in a savagely strong war chapter that saw House Stark overcome huge odds to reclaim their home. Plus, Daenerys got to soar, as her dragons quickly stopped a violent siege with fury and fire." Jeremy Egner of The New York Times also praised the episode: "As directed by Miguel Sapochnik, who also oversaw last season's terrific 'Hardhome' episode, the lengthy sequence"Battle of the Bastards" is the ninth and penultimate episode of the sixth season of HBO's fantasy television series Game of Thrones and its 59th episode overall. It was written by series co-creators David Benioff and D. B. Weiss, and directed by Miguel Sapochnik. Jon Snow and Ramsay Bolton face off in a battle for control of Winterfell. Jon's force, mostly composed of Wildlings, is nearly defeated by the Bolton army, but the latter is overcome when Sansa Stark and Petyr Baelish arrive with the Knights of the Vale. The battle culminates in Ramsay's defeat and capture, followed by his death at the hands of Sansa. Meanwhile, Daenerys Targaryen defeats the Masters in Meereen and begins making new alliances. "Battle of the Bastards" received immense critical acclaim, with several reviewers calling it a "masterpiece", and being praised as one of the series' best episodes as well as one of the greatest television episodes of all time. Critics described the battle in the North as "terrifying, gripping and exhilarating", Harington's performance received high praise, and Daenerys' reunion with her dragons at the beginning of the episode was deemed "thrilling". The eponymous battle took 25 days to film and required 500 extras, 600 crew members, and 70 horses. In the United States, the episode had a viewership of 7.66 million in its initial broadcast. It notably won the Emmy award for Outstanding Directing and Outstanding Writing; the episode was Kit Harington's choice to support his nomination for Outstanding Supporting Actor. Sapochnik also won the Directors Guild of Americasource material. Some of the season's storyline is derived from content not yet published in Martin's A Song of Ice and Fire series, although a significant amount of material from A Feast for Crows, A Dance with Dragons and the upcoming sixth novel The Winds of Winter, which Martin previously outlined to showrunners David Benioff and D.B. Weiss, was used. The season was largely met with positive reviews. The "Battle of the Bastards" episode received immense critical acclaim, with many calling it one of the best television episodes of all time. U.S. viewership rose compared to the previous season, and by approximately 13 percent over its course, from 7.9 million to 8.9 million by the finale. The season won 12 of the 23 Emmy Awards for which it was nominated. Three days before the premiere of the episode "The Red Woman", HBO ordered the seventh season. Due to necessary weather conditions required for filming, the production of the penultimate season of the show was delayed that year. Filming began only on August 31, 2016, at Titanic Studios in Belfast, and ended in February 2017. Unlike previous seasons, the seventh and eighth seasons largely consisted of original content not found in the source material. This season comprised only seven episodes. The showrunners stated that they were unable to produce 10 episodes in the show's usual 12 to 14 month time frame, as Weiss said "It's crossing out of a television schedule into more of a mid-range movie schedule". The average runtime of an episode in this season
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: where did they film hot tub time machine?
-A: Fernie Alpine Resort
-
-Q: who has the right of way in international waters?
-A: Neither vessel
-
-Q: who does annie work for attack on titan?
-A: Marley
-
-Q: when was the immigration reform and control act passed?
-A: November 6, 1986
-
-Q: when was puerto rico added to the usa?
-A: 1950
-
-Q: how many seasons of the bastard executioner are there?
-A:
-==================================================
-
-==================================================
-PROMPT #4:
-==================================================
-the Eagles-Cowboys game. The Eagles then defeated the Minnesota Vikings 26–14 in the first round of the playoffs on January 4, 2009. They defeated their division rival and the defending Super Bowl champions the New York Giants 23–11 in the NFC Divisional Round to advance to their fifth NFC Championship game in nine years. For the third time in a decade, an all-Pennsylvania Super Bowl seemed possible, as the Pittsburgh Steelers won the AFC Championship against the Baltimore Ravens. However, the Eagles lost 32–25 to the Arizona Cardinals. 2009 In August, the team stirred up controversy by signing ex-Falcons quarterback Michael Vick, who had recently been released from prison for federal dogfighting charges. On the regular season opener against Carolina, McNabb sustained a broken rib and sat out the next two games, with Kevin Kolb taking his place as starter. Afterwards, Philadelphia endured a bad loss to New Orleans, but then beat Kansas City the following week. McNabb returned, and Vick also began playing (as he had been barred from the first two games). He got off to a slow start and was used sparingly for much of the season. After a 13–9 loss in Oakland, the Eagles defeated the Redskins 27–17 on Monday Night Football. During this game, running back Brian Westbrook suffered a severe concussion and was knocked unconscious. He returned to play in the Week 10 game against San Diego, but was felled by a second concussion. Afterwards, Philadelphia went on a five-game winning streak and clinching a playoff berth after defeating thedivision, after going 6–0 against their division in 2004. The Eagles became the first team in NFL history to go winless against their division after going undefeated against their division the previous season. After expectations of a return to the Super Bowl, it appeared the Eagles would have to retool in the 2006 off-season to make another run for the Vince Lombardi Trophy. 2006 In 2006, the Eagles started off hot, beating the Houston Texans in Houston 24–10. And they kept the streak going, jumping to a 24–7 lead against the Giants before losing 30–24 in overtime. They won their next two games easily, beating the 49ers in San Francisco and then returning home to limit Brett Favre's Packers to three field goals. The Birds' 3–1 record set the stage for Terrell Owens' return to Philadelphia. Owens' return to Philadelphia was being promoted by Fox as the "game of the year". The game had more at stake, because the winner would take the lead in the NFC East. Owens received boos and jokes about his accidental overdose from a sell-out crowd. Lito Sheppard's game-winning interception for a touchdown sealed the game for the Eagles, 38–24. McNabb's season was already considered one of the finest in Eagles history and his career. After the win against Owens' Cowboys, they faced the upstart New Orleans Saints, with critics claiming the winner would be the NFC favorite. The Eagles ended up losing 27–24 on a last-second field goal. The next week at Tampa, the Eagles would once again lose onto the Chicago Bears, the team that Ryan had helped lead to a Super Bowl XX victory as defensive coordinator. The game became known as the "Fog Bowl", due to the inclement weather during the game. The Eagles lost, 20–12. The following two years would see playoff appearances as well, but the team never made it past the first round. This failure was very frustrating to Eagles fans, as the team was widely regarded as among the most talented in the NFL. On offense, the Eagles were led by quarterback Cunningham, one of the most exciting players of his generation; tight end Keith Jackson; and running back Keith Byars. The defense is commonly acknowledged as among the greatest in league history, and as the best to never win a championship. The two 1989 matches with Dallas were known as the Bounty Bowls. Both were won easily by the Eagles (the Cowboys finished 1–15 that year), and were marked by Ryan insulting new Cowboys coach Jimmy Johnson by placing a bounty on their kicker, and by Eagles fans throwing snowballs at him at Veterans Stadium. On November 12, 1990, during a Monday Night Football game at the Vet, the Eagles defeated the Washington Redskins by a score of 28–14, as the defense scored three of the team's four touchdowns. This game, more lopsided than its score would indicate, was subsequently labeled the Body Bag Game in reference to the number of injuries Eagles players inflicted on the Redskins and the physically tough play of the Eagles in
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: where did they film hot tub time machine?
-A: Fernie Alpine Resort
-
-Q: who has the right of way in international waters?
-A: Neither vessel
-
-Q: who does annie work for attack on titan?
-A: Marley
-
-Q: when was the immigration reform and control act passed?
-A: November 6, 1986
-
-Q: when was puerto rico added to the usa?
-A: 1950
-
-Q: when did the eagles win last super bowl?
-A:
-==================================================
-
-==================================================
-PROMPT #5:
-==================================================
-style=""| AAC regular season |- !colspan=12 style=""|American Athletic Conference women's tournament |- !colspan=12 style=""|NCAA women's tournament Rankings WNBA draft Breanna Stewart, Moriah Jefferson and Morgan Tuck were the first, second and third pick of the 2016 WNBA draft. Players from the same college team have never gone 1,2,3 in any draft in any major sport in history. Awards and honors Breanna Stewart won the NCAA Bridgeport Regional Most Outstanding Player earned the Most Outstanding Player of the Final Four at the 2016 NCAA Division I women's basketball tournament for a record fourth consecutive time won the Associated Press Women's College Basketball Player of the Year for the third consecutive year won her second Wade Trophy, presented to the Division I player of the year was awarded the Naismith College Player of the Year for the third consecutive year won the USBWA Women's National Player of the Year for the third consecutive year won the 2016 American Athletic Conference Women's Basketball Player of the Year won the John R. Wooden Award was the recipient of the 2016 Senior CLASS Award was the recipient of the James E. Sullivan Award Geno Auriemma was awarded the Associated Press College Basketball Coach of the Year won the AAC coach of the year Moriah Jefferson won the Nancy Lieberman Award for the second consecutive year won the WBCA Defensive Player of the Year award won the Dawn Staley Award won the AAC Defender of the Year Morgan Tuck won the AAC Sportsmanship Award was named to the all AAC first teamawarded to the Canadian Interuniversity Sport winner in women's basketball Adelaide Lightning won the Australian Women's National Basketball League Championship The WNBA is founded, with eight initial teams. Sheryl Swoopes is the first player signed. The American Basketball League (ABL) formed, a professional basketball league for women in the United States. It lasted two full seasons, and suspended operations in the third. 1997 Tennessee (29-10) won the NCAA Division I women's basketball tournament North Dakota (28-4) won the NCAA Women's Division II Basketball Championship NYU (29-1) won the NCAA Women's Division III Basketball Championship Manitoba Bisons won the Bronze Baby, awarded to the Canadian Interuniversity Sport winner in women's basketball Sydney Flames won the Australian Women's National Basketball League Championship First WNBA draft, with Tina Thompson as the first player selected. The first game is held on 21 June 1997, between the New York Liberty and the Los Angeles Sparks. The Liberty won 67–57. The Houston Comets win the first WNBA Championship. Trent Tucker Rule adopted by WNBA. 1998 Tennessee (39-0) won the NCAA Division I women's basketball tournament North Dakota (31-1) won the NCAA Women's Division II Basketball Championship Washington (MO) (28-2) won the NCAA Women's Division III Basketball Championship Victoria Vikes won the Bronze Baby, awarded to the Canadian Interuniversity Sport winner in women's basketball Adelaide Lightning won the Australian Women's National Basketball League Championship FIBA World Championship for Women Gold—USA Silver—Russia Bronze—Australia The Houston Comets won the WNBA Championship. 1999 Purdue (34-1) won the NCAA Division I women's basketball tournament North Dakota (31-1)to win that award three times) Breanna Stewart was named the Most Outstanding Player of the Final Four (making her the first person to be most outstanding player of the Final Four four times) Lubbock Christian University (35–0) won the NCAA Division II Women's Basketball Championship Thomas More (33–0) won the NCAA Women's Division III Basketball Championship The Los Angeles Sparks won the WNBA Championship. 2017 The Connecticut Huskies women's basketball team obtained the longest winning streak in college basketball (both men's and women's), 111 straight wins, which started with a win against Creighton on December 23, 2014, and continued for 111 games until March 31, 2017, when they were beaten 66–64 on a last second shot in overtime by Mississippi State in the 2017 NCAA Final Four. This streak included an undefeated season in 2015–16. McGill University (25–9) won the Bronze Baby, awarded to the women's basketball champion of the newly renamed U Sports. Sydney Uni Flames won the Australian Women's National Basketball League Championship South Carolina (33–4) won the NCAA Division I women's basketball tournament Ashland (37–0) won the NCAA Women's Division II Basketball Championship Amherst (33–0) won the NCAA Women's Division III Basketball Championship The Minnesota Lynx won the WNBA Championship. 2018 Carleton won the Bronze Baby, awarded to the U Sports women's basketball champion. Townsville Fire won the Australian Women's National Basketball League Championship Notre Dame (35-3) won the NCAA Women's Division I Basketball Championship Central Missouri (30–3) won the NCAA Women's Division II Basketball Championship Amherst (33–0) Line truncated
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: where did they film hot tub time machine?
-A: Fernie Alpine Resort
-
-Q: who has the right of way in international waters?
-A: Neither vessel
-
-Q: who does annie work for attack on titan?
-A: Marley
-
-Q: when was the immigration reform and control act passed?
-A: November 6, 1986
-
-Q: when was puerto rico added to the usa?
-A: 1950
-
-Q: who won last year's ncaa women's basketball?
-A:
-==================================================
-
-==================================================
-PROMPT #6:
-==================================================
-Isle of WightAs an island, the Isle of Wight maintains a culture close to, but distinct from, that of the south of England. A high proportion of the population are now 'overners' rather than locally born, and so with a few notable exceptions it has more often formed the backdrop for cultural events of wider rather than island-specific significance. The Island has inspired many creative works. Local people often seek to defend their real or perceived culture, and local politics is often dictated by a desire to preserve the traditions and habits of the Island. The first creative flowering occurred during the reign of Queen Victoria, under whose patronage the island became a fashionable destination for the gentry. Literature and other media Alfred, Lord Tennyson was made Baron Tennyson, of Aldworth in the County of Sussex and of Freshwater on the Isle of Wight by Queen Victoria in 1884. The poet Algernon Charles Swinburne grew up at Bonchurch, and said in a letter that he had climbed Culver Cliff at 17. He is buried at Bonchurch. The author Maxwell Gray (Mary Gleed Tuttiett) was born in Newport, and a number of her novels, including the best-known, The Silence of Dean Maitland, are set on the island. The isle has been the setting for several novels, including Julian Barnes's utopian novel England, England, and detective thrillers such as The Fallen by Robert Rennick. It also features in John Wyndham's novel The Day of the Triffids and Simon Clark's sequel The Night of the Triffids. The Iranian-born poet Mimi Khalvatithe first county councils were created in 1888. However, a "Home Rule" campaign led to a separate Isle of Wight County Council being established in 1890, and the island has remained independent of mainland local government ever since. The island was, however, part of the "Assize county" of Hampshire. Like inhabitants of many islands, Islanders are fiercely jealous of their real (or perceived) independence, and confusion over the Island's separate status is a perennial source of friction. It was planned to merge the county back into Hampshire as a district in the 1974 local government reform, but a last minute change led to it retaining its county council. However, since there was no provision made in the Local Government Act 1972 for unitary authorities, the Island had to retain a two-tier structure, with a county council and two boroughs, Medina and South Wight. The borough councils were merged with the county council on 1 April 1995 to form a single unitary authority, the Isle of Wight Council, as recommended by the Local Government Commission for England. The only significant present-day administrative link with Hampshire is the police service, which is joint between Hampshire and the Isle of Wight. House of Commons representation The Isle of Wight forms a single constituency of the House of Commons, with an electorate of 113,021 (as of 2019). This is by far the largest electorate in the country. The 2018 proposals from the Boundary Commission for England will split the constituency into two. The constituency was traditionally a battleground between the
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: where did they film hot tub time machine?
-A: Fernie Alpine Resort
-
-Q: who has the right of way in international waters?
-A: Neither vessel
-
-Q: who does annie work for attack on titan?
-A: Marley
-
-Q: when was the immigration reform and control act passed?
-A: November 6, 1986
-
-Q: when was puerto rico added to the usa?
-A: 1950
-
-Q: when did the isle of wight become an island?
-A:
-==================================================
-
-==================================================
-PROMPT #7:
-==================================================
-"Love Yourself" is a 2015 single by Justin Bieber. Love Yourself may also refer to: Music by BTS Love Yourself: Wonder, the music video theme for the group's song "Euphoria" Love Yourself: Her, the group's fifth extended play Love Yourself: Tear, the group's third Korean studio album and sixth overall Love Yourself: Answer, the group's second compilation album BTS World Tour: Love Yourself, the group's third worldwide concert tour Music by other artists Love Yourself (album), by Prudence Liew, 2000 "Love Yourself" (Billy Porter song), 2019 "Love Yourself" (Mary J. Blige song), 2017 "Love Yourself (Kimi ga Kirai na Kimi ga Suki)", by Kat-Tun, 2010 "Love Yourself", a 1996 song by Blueboy "Love Yourself", a 1997 song by Karen Mok from To Be "Love Yourself", a 2019 song by Sufjan Stevens See also Love Myself (disambiguation)third radio single. When talking to Ryan Seacrest, Bieber said "Love Yourself" is "definitely about someone in my past, someone who I don't want to put on blast," he described the song as "cool because so many people can resonate with that because how many women do we bring back that mom doesn't really necessarily like?." Composition and lyrics "Love Yourself" was written by Ed Sheeran, Benny Blanco and Justin Bieber, with production being done by Benny Blanco, who was also responsible for the song's instrumentation and programming. It features lyrical vocals by Bieber and background vocals by Sheeran. According to the sheet music published at Musicnotes.com by Sony/ATV Music Publishing, the song is composed in the key of E major with a moderate tempo of 100 beats per minute and a time signature of . Bieber's vocals range from the note of B3 to B4. It is an acoustic pop, folk and blue-eyed soul song, featuring just vocals and an electric guitar, and a brief flurry of trumpets, which was considered "the most subdued and least electronic of Bieber's new tracks." Lyrically, "Love Yourself" is a kiss-off to a narcissistic ex-lover who did the protagonist wrong, with Bieber singing in a snappy tone while criticizing "a girl for loving herself too much." In the pre-chorus, he sings with a husky tone in the lower registers: "My mama don't like you, and she likes everyone," "in a style that molds well to Sheeran's," according to Rolling Stone Brittany Spanos. "And I [never] like to admit that"Love Yourself" is a song recorded by Canadian singer Justin Bieber for his fourth studio album Purpose (2015). The song was released as a promotional single on November 9, 2015, and as the album's third official single on December 7, 2015. It was written by Ed Sheeran, Benjamin Levin and Bieber, and produced by Levin. A pop song, "Love Yourself" features an electric guitar and a brief flurry of trumpets as its main instrumentation. During the song, Bieber uses a husky tone in the lower registers. Lyrically, the song is a kiss-off to a narcissistic ex-lover who did the protagonist wrong. On the US Billboard Hot 100 and the UK Singles Chart, the song became Bieber's third consecutive number-one, wherein the United States it spent 24 non-consecutive weeks in the top ten (later named the best-performing single of 2016) and was also Bieber's first number one on the Adult Contemporary chart, while in the United Kingdom it spent six weeks at the top. "Love Yourself" topped the charts in fifteen countries, including Australia, Brazil, Canada, New Zealand, and Sweden. "Love Yourself" was nominated for two Grammy Awards: Song of the Year and Best Pop Solo Performance. It was the seventh-best-selling song of 2016 in the US. The music video for the song was released along with the Purpose: The Movement project on November 14, 2015. It features dancers Keone & Mari Madrid doing an interpretive dance in a house. Bieber promoted the song with performances on TV shows, as well as awards shows throughout 2015–16, while
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: where did they film hot tub time machine?
-A: Fernie Alpine Resort
-
-Q: who has the right of way in international waters?
-A: Neither vessel
-
-Q: who does annie work for attack on titan?
-A: Marley
-
-Q: when was the immigration reform and control act passed?
-A: November 6, 1986
-
-Q: when was puerto rico added to the usa?
-A: 1950
-
-Q: love yourself by justin bieber is about who?
-A:
-==================================================
-
-==================================================
-PROMPT #8:
-==================================================
-Charles I (19 November 1600 – 30 January 1649) was King of England, Scotland, and Ireland from 27 March 1625 until his execution in 1649. He was born into the House of Stuart as the second son of King James VI of Scotland, but after his father inherited the English throne in 1603, he moved to England, where he spent much of the rest of his life. He became heir apparent to the kingdoms of England, Scotland, and Ireland in 1612 upon the death of his elder brother, Henry Frederick, Prince of Wales. An unsuccessful and unpopular attempt to marry him to the Spanish Habsburg princess Maria Anna culminated in an eight-month visit to Spain in 1623 that demonstrated the futility of the marriage negotiation. Two years later, he married the Bourbon princess Henrietta Maria of France. After his succession in 1625, Charles quarrelled with the English Parliament, which sought to curb his royal prerogative. He believed in the divine right of kings, and was determined to govern according to his own conscience. Many of his subjects opposed his policies, in particular the levying of taxes without parliamentary consent, and perceived his actions as those of a tyrannical absolute monarch. His religious policies, coupled with his marriage to a Roman Catholic, generated antipathy and mistrust from Reformed religious groups such as the English Puritans and Scottish Covenanters, who thought his views too Catholic. He supported high church Anglican ecclesiastics and failed to aid continental Protestant forces successfully during the Thirty Years' War. His attempts to forcethe Queen as the supreme governor of the Church of England. After the Spanish Armada was defeated in 1588, Its two levels of administration were the House of Lords, composed of influential peers of the realm and Lords Spiritual, and the House of Commons, which consisted of representative members of the aristocracy and the middle-class. Puritans in the House of Commons began demanding more rights, but their demands were ignored. James I would later have problems with them. Before 1600, the Crown founded corporations including the East India Company to monopolise trade routes. Under her successor, James I, further companies were created to colonise North America, including the London Company and the Virginia Company in 1606, and the Massachusetts Bay Company in 1628. Many religious dissidents left England to settle the new world. Stuart dynasty While Elizabeth I maintained a protestant Church, under her successor James, who unified the Scottish and English Crowns, religious and political tensions grew as he asserted a divine right of Kings. In 1605, Guy Fawkes attempted to blow up Parliament, but was caught, tortured and executed. A wave of repression against catholics followed. James acceded to Puritan requests by commissioning the "King James Bible" in 1604, an English language translation and interpretation of the Bible completed in 1611. Possibly persuaded by his (secretly Catholic) wife, James exempted Catholics from paying tithes to the Anglican Church, but this caused a great decrease in Anglican Church revenue, so he quickly took those rights away. The assertion of divine right prompted a series ofThe concept of "British history" began to emerge in the 1600s, largely thanks to the attempts of King James II to assert that the Union of the Crowns of 1603 had created a Kingdom of Great Britain, which in fact did not come into existence until a century later. The governance of the Kingdom of England and the Kingdom of Scotland remained separate until 1707, and until then in most ways the Scots were excluded from sharing in the English overseas possessions. This page presents a timeline of events in English and Scottish history from 1600 until 1699. 1603 – Death of Queen Elizabeth I on 24 March 1603 England – James VI of Scotland crowned King of England (as James I of England) 1603 England – Plague 1605 England and Scotland – on 5 November, the Gunpowder plot is uncovered, in which Guy Fawkes and other catholic associates attempted to blow up the king, James VI and I and the Parliament of England. 1618 England – Execution of Sir Walter Raleigh 1625 England and Scotland – Death of James VI and I on 27 March 1639 England and Scotland – At war until 1644 in what become known as the Bishops' Wars 1640 England – The Long Parliament summoned. 1642 England – English Civil War begins (see Timeline of the English Civil War) 1652 England – Tea arrives in Britain 1666 England – The Great Fire of London ravages the city, 2–5 September. 1688 England – The Glorious Revolution replaces James II with William III
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: where did they film hot tub time machine?
-A: Fernie Alpine Resort
-
-Q: who has the right of way in international waters?
-A: Neither vessel
-
-Q: who does annie work for attack on titan?
-A: Marley
-
-Q: when was the immigration reform and control act passed?
-A: November 6, 1986
-
-Q: when was puerto rico added to the usa?
-A: 1950
-
-Q: who was the ruler of england in 1616?
-A:
-==================================================
-
-==================================================
-PROMPT #9:
-==================================================
-Le mod Hot Coffee (« café chaud » en anglais) est un mini-jeu de la version de 2004 du jeu vidéo Grand Theft Auto: San Andreas. Normalement indisponible, le mini-jeu est parvenu à la connaissance du public en 2005 par la création du mod Hot Coffee, donnant accès au mini-jeu pour la version destinée aux ordinateurs personnels utilisant le système d'exploitation Windows. Le mini-jeu présente des relations sexuelles entre le personnage principal, Carl Johnson, et sa petite amie Denise. Le nom du mod provient de l'offre faite par la petite amie au personnage principal de venir prendre un « café » chez elle, un euphémisme pour parler de l'autre offre. Bien que le mini-jeu ait été seulement souligné lors de la lancée du mod le et qu'il ait été complètement retiré depuis, des versions du mini-jeu ont été retrouvées sur les versions du jeu développées pour la PlayStation 2 et la Xbox. À la mi-, le mini-jeu soulève une forte controverse politique et judiciaire, ce qui amène le reclassement du jeu en un et son retrait de certaines étagères. Depuis, une version mise à jour de San Andreas, avec le mini-jeu complètement retiré, est sortie, ce qui a permis de ramener le jeu à son classement initial. Un patch, désactivant le mini-jeu et mettant fin au jeu lorsque l'on tente d'y accéder, a également été créée pour la version originale. Notes et références Mod de jeu vidéo Grand Theft Auto Controverse autour du jeu vidéode San Andreas. A ESRB alterou o rótulo do jogo para "Adults Only" (AO) — somente para adultos —, fazendo com que muitos varejistas o retirassem da venda, enquanto que a OFLC emitiu uma classificação recusada, banindo o jogo na Austrália. A Rockstar fez um retalho de todas as cópias de varejo de San Andreas, lançou uma nova versão que bloqueava o acesso ao "Hot Coffee" no final de 2005, e lançou um patch para os proprietários existentes para impedir o acesso. O código-fonte do "Hot Coffee" permanece em todas as versões de San Andreas, mas exige mais esforço para ser restaurado devido à falta de modelos e animações. "Hot Coffee" levantou preocupações de legisladores relacionados às classificações obrigatórias do jogo, e vários processos civis foram iniciados contra a Rockstar. A empresa-mãe da Rockstar, Take-Two Interactive, já estava sob investigação pela Comissão de Valores Mobiliários dos Estados Unidos (SEC) em relação ao comércio de informações privilegiadas, e as ações da legisladora devido ao "Hot Coffee" desencadearam uma investigação separada pela Federal Trade Commission. Esses eventos levaram a Take-Two a perder mais de 163 milhões de dólares em 2006 e fez com que os acionistas demitissem o fundador e presidente da empresa, Ryan Brant, e o substituíssem por Strauss Zelnick. Jogabilidade Em Grand Theft Auto: San Andreas, o jogador, como o protagonista Carl "CJ" Johnson, progride no jogo de ação-aventura por meio de uma série de missões baseadas em uma história. Entre as missões, o jogador pode se envolver em uma série de atividades paralelas para ganhar"Hot Coffee" is the unofficial name for a minigame in the 2004 action-adventure video game Grand Theft Auto: San Andreas by Rockstar Games. While it was not playable in the official game release, the modding community discovered hidden code that, when enabled, allows protagonist Carl "CJ" Johnson to have animated sexual intercourse with his in-game girlfriend. Rockstar Games president Sam Houser wanted to include more role-playing elements in San Andreas while also pushing the Grand Theft Auto series' controversial reputation. The development team was forced to curtail the nudity and sexual content of Houser's original vision, however, to obtain a "Mature" rating from the Entertainment Software Rating Board (ESRB). Rather than removing the content, the developers made it inaccessible to players. Modders discovered the code on the game's PlayStation 2 release, and when San Andreas was released for Windows, modder Patrick Wildenborg disabled the controls around the code. He released this modified code online under the name "Hot Coffee". The discovery of the "Hot Coffee" minigame resulted in intense legal backlash for Rockstar Games and their parent company, Take-Two Interactive. While both companies remained mostly silent on the matter, Rockstar Games released a statement claiming that modders were responsible for the minigame. The ESRB re-rated the game "Adults Only" after an investigation, while the game was banned entirely in Australia until the explicit content was removed. Rockstar Games and Take-Two received a warning from the Federal Trade Commission (FTC) for failing to disclose the extent of graphic content present in the game, while a class action
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: where did they film hot tub time machine?
-A: Fernie Alpine Resort
-
-Q: who has the right of way in international waters?
-A: Neither vessel
-
-Q: who does annie work for attack on titan?
-A: Marley
-
-Q: when was the immigration reform and control act passed?
-A: November 6, 1986
-
-Q: when was puerto rico added to the usa?
-A: 1950
-
-Q: what is the hot coffee mod in san andreas?
-A:
-==================================================
-
-==================================================
-PROMPT #10:
-==================================================
-rate of 802.11ad is 7 Gbit/s. IEEE 802.11ad is a protocol used for very high data rates (about 8 Gbit/s) and for short range communication (about 1–10 meters). TP-Link announced the world's first 802.11ad router in January 2016. The WiGig standard is not too well known, although it was announced in 2009 and added to the IEEE 802.11 family in December 2012. 802.11af IEEE 802.11af, also referred to as "White-Fi" and "Super Wi-Fi", is an amendment, approved in February 2014, that allows WLAN operation in TV white space spectrum in the VHF and UHF bands between 54 and 790 MHz. It uses cognitive radio technology to transmit on unused TV channels, with the standard taking measures to limit interference for primary users, such as analog TV, digital TV, and wireless microphones. Access points and stations determine their position using a satellite positioning system such as GPS, and use the Internet to query a geolocation database (GDB) provided by a regional regulatory agency to discover what frequency channels are available for use at a given time and position. The physical layer uses OFDM and is based on 802.11ac. The propagation path loss as well as the attenuation by materials such as brick and concrete is lower in the UHF and VHF bands than in the 2.4 GHz and 5 GHz bands, which increases the possible range. The frequency channels are 6 to 8 MHz wide, depending on the regulatory domain. Up to four channels may be bonded in either one or two contiguous blocks. MIMO operation isa higher throughput, due to the additional data stream. Data rates Assuming equal operating parameters to an 802.11g network achieving 54 megabits per second (on a single 20 MHz channel with one antenna), an 802.11n network can achieve 72 megabits per second (on a single 20 MHz channel with one antenna and 400 ns guard interval); 802.11n's speed may go up to 150 megabits per second if there are not other Bluetooth, microwave or Wi-Fi emissions in the neighborhood by using two 20 MHz channels in 40 MHz mode. If more antennas are used, then 802.11n can go up to 288 megabits per second in 20 MHz mode with four antennas, or 600 megabits per second in 40 MHz mode with four antennas and 400 ns guard interval. Because the 2.4 GHz band is seriously congested in most urban areas, 802.11n networks usually have more success in increasing data rate by utilizing more antennas in 20 MHz mode rather than by operating in the 40 MHz mode, as the 40 MHz mode requires a relatively free radio spectrum which is only available in rural areas away from cities. Thus, network engineers installing an 802.11n network should strive to select routers and wireless clients with the most antennas possible (one, two, three or four as specified by the 802.11n standard) and try to make sure that the network's bandwidth will be satisfactory even on the 20 MHz mode. Data rates up to 600 Mbit/s are achieved only with the maximum of four spatial streams using one 40maximum net data rate of 54 Mbit/s, plus error correction code, which yields realistic net achievable throughput in the mid-20 Mbit/s. It has seen widespread worldwide implementation, particularly within the corporate workspace. Since the 2.4 GHz band is heavily used to the point of being crowded, using the relatively unused 5 GHz band gives 802.11a a significant advantage. However, this high carrier frequency also brings a disadvantage: the effective overall range of 802.11a is less than that of 802.11b/g. In theory, 802.11a signals are absorbed more readily by walls and other solid objects in their path due to their smaller wavelength, and, as a result, cannot penetrate as far as those of 802.11b. In practice, 802.11b typically has a higher range at low speeds (802.11b will reduce speed to 5.5 Mbit/s or even 1 Mbit/s at low signal strengths). 802.11a also suffers from interference, but locally there may be fewer signals to interfere with, resulting in less interference and better throughput. 802.11b The 802.11b standard has a maximum raw data rate of 11 Mbit/s (Megabits per second) and uses the same media access method defined in the original standard. 802.11b products appeared on the market in early 2000, since 802.11b is a direct extension of the modulation technique defined in the original standard. The dramatic increase in throughput of 802.11b (compared to the original standard) along with simultaneous substantial price reductions led to the rapid acceptance of 802.11b as the definitive wireless LAN technology. Devices using 802.11b experience interference from other products operating in the 2.4
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: where did they film hot tub time machine?
-A: Fernie Alpine Resort
-
-Q: who has the right of way in international waters?
-A: Neither vessel
-
-Q: who does annie work for attack on titan?
-A: Marley
-
-Q: when was the immigration reform and control act passed?
-A: November 6, 1986
-
-Q: when was puerto rico added to the usa?
-A: 1950
-
-Q: what is the maximum data rate for the 802.11a standard select one?
-A:
-==================================================
-
-==================================================
-PROMPT #11:
-==================================================
-Karnataka es un estado de la República de la India. Su capital es Bangalore. Está ubicado al suroeste del país, limitando al noroeste con Goa, al norte con Maharastra, al noreste con Telangana, al este con Andhra Pradesh, al sureste con Tamil Nadu al sur con Kerala, y al oeste con el Mar Arábigo (océano Índico). Con 61 095 297 habitantes en 2011 es el octavo estado más poblado —por detrás de Uttar Pradesh, Maharastra, Bihar, Bengala Occidental y Madhya Pradesh, Tamil Nadu y Rajastán— y con 191 791 km², el séptimo más extenso, por detrás de Rajastán, Madhya Pradesh, Maharastra, Uttar Pradesh, Jammu y Cachemira y Guyarat. Antes de 1973 se le conocía como Estado de Mysore y estaba formado por el antiguo reino de Mysore. En 1956, se incluyeron también algunas zonas anexas en las que el idioma principal era el canarés. Aparte de la capital, otras ciudades importantes son Mysore, Mangalore, Hubli-Dharwad, Bellary y Belgaum. El idioma oficial es el canarés. Según datos del censo de 2001, Karnataka tenía una población de 55.868.200 habitantes. Se cree que el origen del nombre de Karnataka proviene de la unión de los términos karu y nadu que significan "tierra elevada". Historia La prehistoria de Karnataka se remonta a las culturas de hacha de mano del Paleolítico. Herramientas y otros objetos achelenses han sido encontrados en Isampur, valle del Hunsgi, que podrían datar de hace 1,27 millones de años. Hachas paleolíticas de mano y cuchillas en forma de canto rodado, hechas con cuarzo y cuarcita se hanhas his official residence in Vijayawada Chhattisgarh: Raipur is the administrative and legislative capital, while the high court (judiciary capital) is located in Bilaspur. The proposed future capital is Nava Raipur. Jammu and Kashmir: Srinagar serves as the summer capital of the state while Jammu is the winter capital. Every six months, the entire state machinery shifts from one city to another. Kerala: Thiruvananthapuram is the administrative and legislative capital of the state, while the high court is located in Ernakulam. Himachal Pradesh: Shimla is the primary capital city. Dharamshala, which is also the headquarters of the Central Tibetan Administration, is the second winter capital of the state. Madhya Pradesh: Bhopal is the administrative and legislative capital of the state, while the high court is located in Jabalpur. Punjab and Haryana: Both states share Chandigarh as their capital city. The city itself is administered as a Union territory. Odisha: Bhubaneswar is the administrative and legislative capital of the state, while the high court is located in Cuttack. Rajasthan: Jaipur is the administrative and legislative capital of the state, while the high court is located in Jodhpur. Uttarakhand: Dehradun is the administrative and legislative capital, while the high court is located in Nainital. The proposed future capital is Gairsain. Ladakh: Leh and Kargil serve as joint capitals of the Union Territory. South Korea: Seoul remains as the capital and seat of the government's branches, but many government agencies have moved to Sejong City. Malaysia: Kuala Lumpur is the constitutional capital, home of the King, and seat ofstates and union territories located in North India are Hindi, Urdu, Punjabi, Kashmiri, Dogri, and English. Definitions Different authorities and sources define North India differently. Government of India definitions The Northern Zonal Council is one of the advisory councils, created in 1956 by the States Reorganisation Act to foster interstate co-operation under the Ministry of Home Affairs, which included the states of Chandigarh, Delhi, Haryana, Himachal Pradesh, Jammu and Kashmir, Ladakh, Punjab and Rajasthan. The Ministry of Culture established the North Culture Zone in Patiala, Punjab on 23 March 1985. It differs from the North Zonal Council in its inclusion of Uttarakhand and the omission of Delhi. In contrast, the Geological Survey of India (part of the Ministry of Mines) included Uttar Pradesh and Delhi in its Northern Region, but excluded Rajasthan and Chandigarh, with a regional headquarters in Lucknow. Wider definition Indian press definition The Hindu newspaper puts Bihar, Delhi and Uttar Pradesh related articles on its North pages. Articles in the Indian press have included the states of Bihar, Gujarat, Jharkhand, Madhya Pradesh, and West Bengal in North India as well. Latitude-based definition The Tropic of Cancer, which divides the temperate zone from the tropical zone in the Northern Hemisphere, runs through India, and could theoretically be regarded as a geographical dividing line in the country. Indian states that are entirely above the Tropic of Cancer are Himachal Pradesh, Punjab, Haryana, Delhi, Uttarakhand, Uttar Pradesh and Bihar and most of North East Indian states. However that definition would also include major parts of Rajasthan,
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: where did they film hot tub time machine?
-A: Fernie Alpine Resort
-
-Q: who has the right of way in international waters?
-A: Neither vessel
-
-Q: who does annie work for attack on titan?
-A: Marley
-
-Q: when was the immigration reform and control act passed?
-A: November 6, 1986
-
-Q: when was puerto rico added to the usa?
-A: 1950
-
-Q: which state is located in the centre of india?
-A:
-==================================================
-
-==================================================
-PROMPT #12:
-==================================================
-"Runaway Child, Running Wild" (shown as "Run Away Child, Running Wild" on the label of the original single) is a 1969 hit single for the Gordy (Motown) label, performed by The Temptations and produced by Norman Whitfield. The single was both the second from their landmark Cloud Nine LP, and the second of their "psychedelic soul" tracks penned by Whitfield and former Motown artist Barrett Strong. Featuring all five Temptations - Dennis Edwards, Eddie Kendricks, Paul Williams, Melvin Franklin, and Otis Williams - trading verses and harmony lines, "Runaway Child, Running Wild" paints a tale of a young boy (presumably a preteen) who runs away from home after being punished for playing hooky i.e. absence from school without permission. The boy wanders the dark streets alone, eventually realizing he cannot survive on his own, but cannot find his way home, and ends up lost, frightened by strangers, unfamiliar landmarks, and his own thoughts. "Runaway child runnin' wild," the Temptations tell the boy during the chorus, "you better go back home/where you belong". The Temptations alternately express and depict his fears, with the tension of the record building to a climax over the first five minutes of the record. At this point, the vocals fade out, and the record briefly gives voice to its young protagonist, who cries desperately for his mother before fading into the mix. "Runaway Child" then segues into an extended instrumental passage, during which Earl Van Dyke's Hammond organ, Joe Messina's electric guitar, and Dennis Coffey's distorted wah-wah pedal guitar take center stageAll the Way Home may refer to: All the Way Home (play), a 1960 play by Tad Mosel adapted from the James Agee novel A Death in the Family All the Way Home (1963 film), a 1963 film adapted from Mosel's play and Agee's novel All the Way Home (1981 film), a TV adaptation of the play and novel, starring Sally Field and Ellen Corby All the Way Home (novel), a 2001 novel by Patricia Reilly Giff Music All the Way Home, a 2012 solo album by Cathy Jordan All The Way Home, a 1970 album by Randy & Gary Scruggs "All the Way Home" (Bruce Springsteen song) "All the Way Home" (Tamar Braxton song), 2013 "All the Way Home", a song recorded by Frank Sinatra written by Teddy Randazzo "All the Way Home", a 1959 song by Andy Williams released on Andy Williams Sings Steve Allen "All the Way Home", a 1960 song by Bobby Darin released on For Teenagers Only "All the Way Home", a 1992 song by Spinal Tap released on Break Like the Wind "All the Way Home", a 1994 song by Third Matinee released on Meanwhile "All the Way Home", a 2000 song by Andrew Peterson released on Carried Along "All the Way Home", a 2000 song by Peter Mulvey released on The Trouble with Poets "All the Way Home", a 2004 song by Jerry Gaskill released on Come Somewhere "All the Way Home", a 2014 song by Melissa Etheridge released on This Is M.E. See also"Sorry (I Ran All the Way Home)" is a song written by Artie Zwirn and Harry Giosasi and produced and arranged by LeRoy Holmes. The single was performed by New York-based doo-wop group The Impalas. It reached #2 on the U.S. pop chart, behind both The Happy Organ by Dave "Baby" Cortez and Kansas City by Wilbert Harrison. It also went to #14 on the U.S. R&B chart. Overseas, "Sorry (I Ran All the Way Home)" went to #28 on the UK Singles Chart in 1959. The song was featured on The Impalas' 1959 album, Sorry (I Ran All the Way Home). The song ranked #24 on Billboard's Year-End top 100 singles of 1959. Other versions Guy Darrell and The Midniters, as a single in 1964, but it did not chart. Heinz, on his 1964 album, Tribute to Eddie. The Royal Showband Waterford, as the B-side to their 1964 single "Huckle Buck". Phil Orsi and The Little Kings, as a single in 1966, but it did not chart. References 1959 songs 1959 singles 1964 singles 1966 singles Doo-wop songs
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: where did they film hot tub time machine?
-A: Fernie Alpine Resort
-
-Q: who has the right of way in international waters?
-A: Neither vessel
-
-Q: who does annie work for attack on titan?
-A: Marley
-
-Q: when was the immigration reform and control act passed?
-A: November 6, 1986
-
-Q: when was puerto rico added to the usa?
-A: 1950
-
-Q: who sang i ran all the way home?
-A:
-==================================================
-
-==================================================
-PROMPT #13:
-==================================================
-Wallace is a Scottish & English given name meaning 'Wales' or 'of Wales, and may refer to: Wallace (footballer, born 1989), Brazilian footballer, most recently played for Rapid Bucharest Wallace (footballer, born May 1994), Brazilian footballer Wallace (footballer, born October 1994), Brazilian footballer, plays for Lazio Wallace (footballer, born 1986), Brazilian footballer Wallace W. Andrew (1850–1919), American politician Wallace Barnes (1926–2020), American politician Wallace Beery (1885–1949), American actor Wallace "Wally" Berman (1926–1976), American filmmaker and artist Wallace Carothers (1896–1937), inventor of nylon Wallace Chung (born 1974), Hong Kong singer and actor Wallace E. Conkling (1896–1979), American Episcopal bishop Wallace de Souza (born 1987), Brazilian volleyball player generally known simply as "Wallace" Wallace H. Graham (1910–1996), White House Physician 1945–1953 Wallace Wilson Graham (1815–1898), American lawyer and legislator from Wisconsin Wallace Gustafson (1925–2018), American lawyer and politician Wallace Johnson (baseball) (born 1956), American former professional baseball player and coach Wallace E. Johnson (1901–1988), co-founder of Holiday Inn Wallace F. Johnson (1889–1971), American tennis player Wallace J.S. Johnson (1913–1979), Mayor of Berkeley, California during the 1960s Wallace Reid (1891–1923), American actor during the silent-film era Wallace M. Rogerson, American exercise leader Wallace Arthur Sabin (1869–1937), composer and organist Wallace Clement Sabine (1868–1919), founder of acoustic science Wallace Shawn (born 1943), American actor, comedian and playwright Wallace Stegner (1909–1993), American environmentalist and fiction writer Wallace Stevens (1879–1955), American poet Wallace Thayer (1866–1944), New York assemblyman Wallace Thompson (1896–1952), American lawyer and legislator Wallace "Wally" Wolf (1930–1997), American swimmer, water polo player, and Olympic champion Wallace "Wally" Wingert (born 1961),Wallace, was an Ulster-Scots immigrant from the village Kilrea in County Londonderry, Ireland, who arrived in Philadelphia in 1823. May (née Broadhead) was born in New York City but was raised by an aunt in Muscatine, Iowa, after her parents' death. Wallace's family moved to Ames, Iowa, in 1892 and to Des Moines, Iowa, in 1896. In 1894, the Wallaces established an agricultural newspaper, Wallace's Farmer. It became extremely successful and made the family wealthy and politically influential. Wallace took a strong interest in agriculture and plants from a young age and befriended African-American botanist George Washington Carver, with whom he frequently discussed plants and other subjects. Wallace was particularly interested in corn, Iowa's key crop. In 1904, he devised an experiment that disproved agronomist Perry Greeley Holden's assertion that the most aesthetically pleasing corn would produce the greatest yield. Wallace graduated from West High School in 1906 and enrolled in Iowa State College later that year, majoring in animal husbandry. He joined the Hawkeye Club, a fraternal organization, and spent much of his free time continuing to study corn. He also organized a political club to support Gifford Pinchot, a Progressive Republican who was head of the United States Forest Service. Journalist and farmer Wallace became a full-time writer and editor for Wallace's Farmer after graduating from college in 1910. He was deeply interested in using mathematics and economics in agriculture and learned calculus as part of an effort to understand hog prices. He also wrote an influential article with pioneering statistician George W. SnedecorThe Clan Wallace is a Lowlands Scottish Clan and is officially recognized as such by the Lord Lyon King of Arms. The most famous member of the clan was the Scottish patriot William Wallace of the late 13th and early 14th centuries. History Origins of the clan The Wallace family first came to Scotland with a Breton family in the 11th century. David I of Scotland was eager to extend the benefits of Norman influence and gave grants to the nobles of the south. Among them was Walter fitz Alan, who the Scottish king appointed his Steward in 1136. One of Fitzallan's followers was Richard Wallace from Oswestry who came north to try to improve his fortunes. Oswestry is on the Welsh border. So it is possible that the name Wallace may be a corruption of Le Waleis meaning the "Welshman". One similar theory is the name's being derived from the Old English wylisc (pronounced "wullish"), meaning "foreigner" or "Welshman".. However, while it is possible that the Wallaces in the Clyde area were originally Britons from Wales, who came north with David I of Scotland in the eleventh century, another theory is that they were Britons who had settled in Strathclyde/Strathclyde kingdom in the tenth century, as the “Welsh” term was also used for the Cumbric-speaking Celtic Britons of that area. The SLine truncated
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: where did they film hot tub time machine?
-A: Fernie Alpine Resort
-
-Q: who has the right of way in international waters?
-A: Neither vessel
-
-Q: who does annie work for attack on titan?
-A: Marley
-
-Q: when was the immigration reform and control act passed?
-A: November 6, 1986
-
-Q: when was puerto rico added to the usa?
-A: 1950
-
-Q: where did the last name wallace come from?
-A:
-==================================================
-
-==================================================
-PROMPT #14:
-==================================================
-that episode, Robinette accuses McCoy and the D.A.'s office of participating in a giant conspiracy to make every Muslim appear to be a terrorist. However, he maintains a respectful relationship with the DA's office, even sometimes having drinks with his now-opponents. In 2017, the character appeared in "Uncertainty Principle", an episode of Chicago P.D., which is set in the same fictional universe as Law & Order. In the episode, Robinette travels to Chicago to defend CPD Officer Kevin Atwater (LaRoyce Hawkins) at the request of Sergeant Hank Voight (Jason Beghe), and goes up against Ben Stone's son, Cook County Assistant State's Attorney Peter Stone (Philip Winchester). References Law & Order characters Fictional African-American people Fictional assistant district attorneys Fictional characters from New York City Television characters introduced in 1990 Fictional lawyers Crossover characters in television American male characters in television pt:Paul Robinettetime has already declared himself "unavailable" to hear the case. In the Chicago Justice episode "Uncertainty Principle", Stone receives a call from his son, Peter, who had just faced off against his former colleague Paul Robinette. The opening scene of the Law & Order: Special Victims Unit episode "The Undiscovered Country" (aired in 2018) reveals that Stone has died. His successor, Jack McCoy, gives the eulogy at the funeral. The Special Victims Unit episode "Dear Ben" mentions Stone several times. The plot of the episode revolves around the cold case of a serial rapist whom Stone repeatedly attempted to track down and prosecute, to no avail. The Special Victims Unit detectives eventually bring the rapist to justice. Family Stone's son is Peter Stone (Philip Winchester), formerly a Cook County Assistant State's Attorney and lead character on Chicago Justice. Peter first appears on the Chicago P.D. episode "Justice", which is the backdoor pilot for Chicago Justice. Peter also appears in season 19 of Law & Order: Special Victims Unit, and eventually becomes the titular sex crimes bureau's ADA. In several episodes, Peter describes Ben as a distant father who neglected his family in favor of work. Stone also had a daughter, Pamela (Amy Korb), who is introduced in the SVU episode "Send in the Clowns". Pamela has paranoid schizophrenia, and has been hospitalized for most of her life. According to Peter, their father would visit Pamela once a week, and Peter has continued the visits since their father's death. She is murdered by sex traffickers in theBenjamin "Ben" Stone is a fictional character portrayed by Michael Moriarty in the TV drama Law & Order. He was the Executive Assistant District Attorney for New York County until his resignation at the end of season four. He appeared in 88 episodes. Character overview Stone works in the Manhattan District Attorney's office under Alfred Wentworth (Roy Thinnes, in the pilot episode "Everybody's Favorite Bagman") and Adam Schiff (Steven Hill). He was born in 1944 and raised in an Irish Catholic family. He was raised mostly by his Irish grandmother, his father having been an alcoholic. Stone is divorced and has a son, Peter Stone (Philip Winchester), who was the Deputy Chief of the Special Prosecutions Bureau in Chicago and was the Chief ADA of the Sex Crimes Bureau in New York, and a daughter, Pamela (Amy Korb). Stone's prosecutorial methods are portrayed as being grounded in moral principles stemming from his Catholic faith. He is shown to be anti-abortion and opposed to the death penalty. He also is a strong advocate of social justice, having marched in the civil rights movement. It is implied that Stone, like actor Moriarty, is a Dartmouth College alumnus, with references to "fraternity row" and New Hampshire. Stone is also likely a graduate of New York University Law School, evidenced by the diplomas on his wall in several episodes. Career within the show Stone became Executive Assistant DA in 1985 after convicting con artist and murderer Philip Swann (Željko Ivanek), who years later enters a civil suit against Stone when
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: where did they film hot tub time machine?
-A: Fernie Alpine Resort
-
-Q: who has the right of way in international waters?
-A: Neither vessel
-
-Q: who does annie work for attack on titan?
-A: Marley
-
-Q: when was the immigration reform and control act passed?
-A: November 6, 1986
-
-Q: when was puerto rico added to the usa?
-A: 1950
-
-Q: who was the actor that played ben stone on law and order?
-A:
-==================================================
-
-==================================================
-PROMPT #15:
-==================================================
-of The Lion Guard. Nala is the most significant female character in The Lion King. As the film was inspired by William Shakespeare's tragedy Hamlet, Nala is considered to be The Lion King's equivalent of Hamlet's love interest Ophelia, although differences remain between the two characters. Many early concepts first developed for Nala were eventually abandoned, including a brother and father for the character, as well as her being romantically pursued by Scar. While critical reception towards Nala has been generally mixed, Kelly's vocal performance has been praised. Nala appears in the Broadway musical adaptation of the film, first played by singer Heather Headley. The character also appears in the television series The Lion Guard, with Gabrielle Union replacing Kelly as the voice of Nala. Beyoncé and Shahadi Wright Joseph voice the character in the 2019 CGI live adaptation of the original film directed by Jon Favreau. Development Conception and creation Screenwriter Linda Woolverton conceived Nala "as part of a gradual progression ... which have driven recent public conversation about what young girls should be able to expect from their cinematic role models." In early versions of the screenplay, Nala has a younger brother named Mheetu, who enjoys accompanying her and her best friend Simba on their adventures. Pronounced "me too", the character's name is a reference to this personality trait. Simba was intended to save the cub from a wildebeest stampede, and Nala would eventually become responsible for protecting Mheetu from Simba's tyrannical uncle Scar. Nala also had a fox friend named Bhati. According toNala is a fictional character in Disney's The Lion King film franchise. Introduced in the animated film The Lion King (1994), Nala subsequently appears as a less prominent character in the film's sequels The Lion King II: Simba's Pride (1998) and The Lion King 1½ (2004), and serves as a recurring character in The Lion Guard (2015–2019). In the original film, the adult Nala is voiced by American actress Moira Kelly. Young Nala's speaking voice is provided by actress Niketa Calame, while singers Laura Williams and Sally Dworsky provide the singing voices of young and adult Nala respectively. Nala is introduced as the daughter of an unnamed lion and Sarafina, the best friend of Simba, and ultimately becomes his wife as well as the daughter-in-law of Mufasa and Sarabi and the niece-in-law of Scar by the end of The Lion King. Nala becomes Simba’s wife as well as his Queen. Nala is also Kopa, Kiara and Kion’s mother. Several years after Simba's uncle Scar has killed Simba's father Mufasa and taken the throne, Nala desperately ventures into the jungle to find help. Upon unexpectedly reuniting with an adult Simba, who she had long been tricked by Scar into presuming dead, Nala encourages him to return to Pride Rock, overthrow his uncle and ultimately become king. As Simba's queen, Nala has a son, Kopa, an adventurous cub in The Lion King: Six New Adventures, a daughter, Kiara, whose story is explored in The Lion King: Simba's Pride and another son named Kion, who serves as the protagonist"would grace any red carpet." Kelly's performance as Nala is responsible for introducing the actress to a younger audience. Previously, she had been better known for appearing in more adult-oriented live-action films at that time. In 2011, Kelly revealed that fans of the film rarely recognize her speaking voice as the voice of Nala, although Robert DeSalvo of NextMovie.com claims that the actress "has a ... distinctive voice that The Lion King fans will instantly recognize as the voice of adult Nala." Characterization and themes A straight woman character, Kelly described Nala as a sensible and nurturing yet authoritative character, while Amber Leab of Bitch Flicks described her as strong, independent and intelligent. Occupying the role of the film's female lead, Nala is The Lion King's most important female character, who contributes to the film's "small romance element". Often identified as the film's second most important character, Taylor Orci of The Atlantic felt that "Nala is really the agent of change in The Lion King", dismissing Simba as a "rich, lazy boyfriend." Leab observed that The Lion King's male characters tend to "take the center stage" while "female characters take a backseat to the action." Lenka Křivánková of Masaryk University wrote in her thesis "1990s Hollywood Break-Away Hits: A Feminist Perspective" that she was not particularly surprised by the film's lack of strong roles for women because of its Shakespearean source material, dubbing the film "an old traditional fairy tale with all its traditional features", including patriarchy and monarchism. Writing for the University of Waterloo's Kinema,
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: where did they film hot tub time machine?
-A: Fernie Alpine Resort
-
-Q: who has the right of way in international waters?
-A: Neither vessel
-
-Q: who does annie work for attack on titan?
-A: Marley
-
-Q: when was the immigration reform and control act passed?
-A: November 6, 1986
-
-Q: when was puerto rico added to the usa?
-A: 1950
-
-Q: who does the voice of nala in the lion king?
-A:
-==================================================
-
-==================================================
-PROMPT #16:
-==================================================
-Gloria Abbott Bardwell is a fictional character from The Young and the Restless, an American soap opera on the CBS network. The role was originated by Joan Van Ark on April 6, 2004, who portrayed the role until 2005, with Judith Chapman assuming the role on January 17, 2005. She is the mother of Michael Baldwin (Christian LeBlanc) and Kevin Fisher (Greg Rikaart). She is known for attracting troublesome men, and being quite troublesome herself. Gloria can be naive at times, always taking things to the extreme. Known for her over-the-top personality, she owned and operated Gloworm, the hottest Genoa City club, until it burned down. Casting In March 2004, it was reported that former Knots Landing star Joan Van Ark would join The Young and the Restless as Gloria Fisher, mother of Michael Baldwin and Kevin Fisher. CBS reportedly wanted a "big name" star to tackle the role of Gloria. Van Ark debuted on April 6, 2004. By December, news broke that Van Ark had decided to leave The Young and the Restless because she was unaccustomed to the rigors of daytime television, including the long work days and the fast turnaround on scripts and production. Within weeks, the role was recast with Judith Chapman; Van Ark was last seen on January 7, 2005, and Chapman assumed the role on January 17, 2005. In March 2011, it was announced that Chapman had been taken off her contract and bumped to a recurring status. In October 2014, it was announced that Chapman would appear as Gloriaand Love Is a Many Splendored Thing, The Young and the Restless' has been successful at replacing some of its lead characters with other actors. Most often, major characters are played by the same actor for decades; if they left the show, the characters left with them. But in the case of The Young and the Restless, their replacements were often popular and remade the character in their own image. When William Grey Espy left the show in 1975, the role of Snapper Foster was given to then-unknown actor David Hasselhoff. Peter Bergman has won three Emmy Awards after replacing Lester as Jack Abbott. And Jess Walton, who took over the role of Jill Foster Abbott after original cast member Brenda Dickson was fired in 1987 amid a fallout with producers, has earned two Emmys. In 2004, Joan Van Ark joined the cast as Gloria Fisher, Michael Baldwin's mother, remaining until early 2005. She was replaced by Judith Chapman. The role of Lily Winters was made popular by actress Christel Khalil. The Young and the Restless recast Lily with Davetta Sherwood in 2005 but the role was eventually given back to Khalil in 2006. The 2005 the role of Mackenzie Browning was recast from Ashley Bashioum to Rachel Kimsey. Kimsey was released from her contract in May 2006 and the character returned in 2009 played by Clementine Ford. The recasting of Mac took place shortly after the character of Billy Abbott returned, played by Billy Miller. There have been many recasts of the role of Billy1990–1991, 1994, 2008–2017, 2019-present (currently played by Elizabeth Hendrickson) Rick Forrester (The Bold and the Beautiful) - 1990–2006, 2007–present (currently played by Jacob Young) Michael Baldwin (The Young and the Restless) - 1991–1993, 1997–present (played by Christian LeBlanc) Nate Hastings (The Young and the Restless) - 1992–2002, 2011, 2019–present (currently played by Sean Dominic) Billy Abbott (The Young and the Restless) - 1993–2003, 2006, 2008–present (currently played by Jason Thompson) Sharon Newman (The Young and the Restless) - 1994–present (currently played by Sharon Case) Phyllis Summers (The Young and the Restless) - 1994–1997, 1997-1999, 2000–2013, 2014–present (currently played by Michelle Stafford) Adam Newman (The Young and the Restless) - 1995–1997, 2002, 2008–2016, 2019-present (currently played by Mark Grossman) Lily Winters (The Young and the Restless) - 1995–1998, 2002–present (currently played by Christel Khalil) References External links 1927 births 2005 deaths Daytime Emmy Award winners Soap opera producers American soap opera writers Burials at Westwood Village Memorial Park Cemetery United States Navy corpsmen University of Michigan alumni 20th-century American screenwriters United States Navy personnel of World War II Deaths from dementia in California Deaths from Alzheimer's disease
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: where did they film hot tub time machine?
-A: Fernie Alpine Resort
-
-Q: who has the right of way in international waters?
-A: Neither vessel
-
-Q: who does annie work for attack on titan?
-A: Marley
-
-Q: when was the immigration reform and control act passed?
-A: November 6, 1986
-
-Q: when was puerto rico added to the usa?
-A: 1950
-
-Q: who plays gram on the young and the restless?
-A:
-==================================================
-
-==================================================
-PROMPT #17:
-==================================================
-The oath of office of the president of the United States is the oath or affirmation that the president of the United States takes upon assuming office. The wording of the oath is specified in Article II, Section One, Clause 8, of the United States Constitution, and a new president must take it before exercising or carrying out any official powers or duties. This clause is one of three oath or affirmation clauses in the Constitution, but it is the only one that actually specifies the words that must be spoken. Article I, Section 3 requires Senators, when sitting to try impeachments, to be "on Oath or Affirmation." Article VI, Clause 3, similarly requires the persons specified therein to "be bound by oath or affirmation, to support this Constitution." The presidential oath requires much more than that general oath of allegiance and fidelity. This clause enjoins the new president to swear or affirm: "I will to the best of my ability, preserve, protect and defend the Constitution of the United States." Text Ceremony A newly elected or re-elected president of the United States begins their four-year term of office at noon on the twentieth day of January following the election, and, by tradition, takes the oath of office during an inauguration on that date; prior to 1937 the president's term of office began on March 4. If January 20 falls on a Sunday, the president will be sworn in that day by taking the oath privately, but will then re-take the oath in a public ceremonyHis Majesty Charles III, his heirs, and successors, according to law. Since 1 January 2004, applicants for British citizenship are also required to make a pledge (see Pledges below). United States The United States Oath of Allegiance (officially referred to as the "Oath of Allegiance," 8 C.F.R. Part 337 (2008)) is an oath that must be taken by all immigrants who wish to become United States citizens. I hereby declare, on oath, that I absolutely and entirely renounce and abjure all allegiance and fidelity to any foreign prince, potentate, state, or sovereignty of whom or which I have heretofore been a subject or citizen; that I will support and defend the Constitution and laws of the United States of America against all enemies, foreign and domestic; that I will bear true faith and allegiance to the same; that I will bear arms on behalf of the United States when required by the law; that I will perform noncombatant service in the Armed Forces of the United States when required by the law; that I will perform work of national importance under civilian direction when required by the law; and that I take this obligation freely without any mental reservation or purpose of evasion; so help me God. For people who object to taking an oath (or are not religious), the words "on oath" can be replaced with "and solemnly affirm", and the words "so help me God" can be omitted. Pledges In some countries, a pledge is taken by a new citizen, as opposed to anThe Oath of Allegiance of the United States is the official oath of allegiance that must be taken and subscribed by every immigrant who wishes to become a United States citizen. The oath may be administered by any immigration judge or any authorized officer of the United States Citizenship and Immigration Services (USCIS), including by any eligible federal judge. Text The current Oath of Allegiance of the United States is as follows: Modifications According to U.S. regulations, the phrase "so help me God" is optional and that the words on oath can be substituted with and solemnly affirm. According to U.S. Congress, if the prospective citizen is unable or unwilling to promise to bear arms or perform noncombatant military service because of "religious training and belief", he or she may request to leave out those clauses. The law specifies: Qualifying for this modification may require supporting documentation from the applicant's religious organization. The applicant is not required to belong to a specific religious group, but must have "a sincere and meaningful belief that has a place in the applicant's life that is equivalent to that of a religious belief." Renunciation of title or order of nobility New citizens who hold foreign titles must also renounce those titles by adding the following phrase:I further renounce the title of (give title or titles) which I have heretofore held; or I further renounce the order of nobility (give the order of nobility) to which I have heretofore belonged. Waivers The requirement to take and subscribe to the Oath of
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: where did they film hot tub time machine?
-A: Fernie Alpine Resort
-
-Q: who has the right of way in international waters?
-A: Neither vessel
-
-Q: who does annie work for attack on titan?
-A: Marley
-
-Q: when was the immigration reform and control act passed?
-A: November 6, 1986
-
-Q: when was puerto rico added to the usa?
-A: 1950
-
-Q: what is the oath that new citizens take?
-A:
-==================================================
-
-==================================================
-PROMPT #18:
-==================================================
-leaving him for dead on the banks of a lava flow, where Vader sustains severe burns. Palpatine finds a barely alive Vader and takes him to Coruscant, where his mutilated body is treated and covered in the black suit first depicted in the original trilogy. When Vader asks if Padmé is safe, Palpatine says that he killed her out of anger, devastating him. At the end of the film, Vader supervises the construction of the first Death Star alongside Palpatine and Tarkin (Wayne Pygram). Jake Lloyd played Anakin Skywalker as a child in The Phantom Menace, while Hayden Christensen played him as a young adult in the following two films. James Earl Jones reprised his role as the voice of Vader under his mask. Sequel trilogy Thirty years after the Galactic Civil War, Darth Vader's melted helmet appears in The Force Awakens (2015), in which Vader's grandson Kylo Ren (Adam Driver) – who has followed in his grandfather's footsteps by falling to the dark side and betraying the Jedi – is seen addressing him, though Vader does not appear in the film. At one point, his helmet was considered as the film's MacGuffin. The helmet appears again in The Rise of Skywalker (2019), when Kylo briefly meditates with it, and during the film's first duel between Kylo and Rey (Daisy Ridley). The helmet is last seen on the planet Kijimi, which is later destroyed by a Sith Star Destroyer. The film also reveals that the voice which Ren perceived coming from Vader's helmet in The Forcea sculptor, he was imprisoned at a young age for his creations, which scared most people who saw them. Momin was eventually rescued by a Sith Lady named Shaa, who trained him in the dark side of the Force until he became more powerful than her and killed her. Momin then built a superweapon to destroy a city and perished when the Jedi intervened to stop him, losing control over the energy he wielded and causing his physical body to be destroyed, leaving behind only his mask with his spirit inside. Many years later, the mask was retrieved from the Jedi Archive vault by Darth Sidious, who gave it to Darth Vader as a gift for his journey to Mustafar, which Vader sought to make his personal stronghold. After Momin's spirit killed some of his personnel, Vader examined the mask and learned of Momin's past, before letting him possess the body of a Mustafarian and build a fortress for him. Shortly after Momin finished building of the fortress, Vader was distracted by an invasion of Mustafarians, which Momin took advantage of to open the door to the Dark Side and resurrect himself. However, after challenging Vader to a duel, Momin met a quick demise by being crushed with a giant rock. Despite Momin's death, his dark essence remained in the mask for several more years. At one point, Momin's mask was aboard the Imperial yacht Imperialis, when a group of thieves led by Lando Calrissian attempt to rob the ship. After Momin's spirit possessed the bodiesof the most iconic villains in popular culture, and has been listed among the greatest villains and fictional characters ever. His masked face is one of the most iconic character designs of all time. Originally a slave on Tatooine, Anakin Skywalker is a Jedi prophesied to bring balance to the Force. He is lured to the dark side of the Force by Chancellor Palpatine (Darth Sidious) and becomes a Sith Lord, assuming the name of Darth Vader. After a lightsaber battle with his former mentor Obi-Wan Kenobi on Mustafar, in which he is severely injured, Vader is transformed into a cyborg. Now the Emperor's handpicked right-hand, he serves the Galactic Empire for over two decades as the Commander-in-Chief of its forces, purging the last Jedi and hunting the Rebels. Vader ultimately redeems himself by saving his son, Luke Skywalker, and killing Palpatine, sacrificing his own life in the process. He is also the secret husband of Padmé Amidala, the biological father of Princess Leia, and the grandfather of Kylo Ren (Ben Solo). In the non-canonical Star Wars Legends continuity, he is also the grandfather of Ben Skywalker, his eponym Anakin Solo, Jaina Solo and Darth Caedus (Jacen Solo), and the great-grandfather of Allana Solo. The character has been portrayed by numerous actors: David Prowse physically portrayed Vader while James Earl Jones has voiced him in all of the films and some television shows, and Sebastian Shaw portrayed the unmasked Anakin in Return of the Jedi, as well as the character's spirit in the original release of
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: where did they film hot tub time machine?
-A: Fernie Alpine Resort
-
-Q: who has the right of way in international waters?
-A: Neither vessel
-
-Q: who does annie work for attack on titan?
-A: Marley
-
-Q: when was the immigration reform and control act passed?
-A: November 6, 1986
-
-Q: when was puerto rico added to the usa?
-A: 1950
-
-Q: who is under the mask of darth vader?
-A:
-==================================================
-
-==================================================
-PROMPT #19:
-==================================================
-(4) his wife taken from him, (5) circumcision, (6) Hagar and Ishmael driven away, (7) the binding of Isaac, and (8) buying the land to bury Sarah. Genesis chapter 24 Josephus reported that Rebekah told Abraham's servant, "my father was Bethuel, but he is dead; and Laban is my brother; and, together with my mother, takes care of all our family affairs, and is the guardian of my virginity." In classical rabbinic interpretation The parashah is discussed in these rabbinic sources from the era of the Mishnah and the Talmud: Genesis chapter 23 A Midrash noted that recorded that "the life of Sarah was a hundred and seven and twenty years" rather than "one-hundred-twenty-seven years," and deduced that as the righteous are whole and unblemished by sin, so are their years reported whole in the Bible. Thus the Midrash taught that at the age of 20, Sarah was as at the age of 7 in beauty, and at the age of 100, she was as at the age of 20 in sin (the age below which Providence does not punish for sin). Rabbi Haggai said in Rabbi Isaac's name that all of the Matriarchs were prophets. Rabbi Abba bar Kahana interpreted the words, "The sun rises, and the sun sets," in Ecclesiastes to teach that before God causes the sun of one righteous person to set, God causes the sun of another righteous person to rise. Thus a Midrash taught that before God allowed Sarah's sun to set, God caused Rebekah's sun to rise. Thus firstGabriel gives a prophecy about seventy weeks and the "Anointed One". If one adds the 180 days that Elizabeth was pregnant before Mary's conception plus the 270 days of Mary's pregnancy plus the forty days of "purification" in Luke 2:22, one gets 490 days, or seventy weeks. Many Christians have seen this as a fulfillment of prophecy, but skeptics tend to counter that Luke could be constructing his story to fit what he sees as fulfillments of prophecy, whether real or not. God's promise of a messiah from the house of David is foretold in 2 Samuel 7. It is significant that Luke states that Mary lives in Nazareth, considered a small backwater town. He thus has God's grace bestowed on a young, unmarried woman living in an insignificant town. Luke frequently has favor shown on women, sinners, and various "unimportant" people. Paul never mentions a virgin birth for Jesus, but in Romans speaks of him as the Son of God by virtue of his resurrection, and states he had a more than human nature. Luke here has Jesus' divine nature declared from the first moment of his conception. Mary and Elizabeth (1:39–56) Mary then goes and visits her relative Elizabeth. Upon meeting the pregnant Mary, Elizabeth feels John move in her womb and is "filled with the Holy Spirit" (verse 41). This may be related to . Elizabeth gives praise to Mary: "Blessed are you among women, and blessed is the fruit of your womb!", words which echo Moses' declaration to the people of Israelthan a stillborn child, which at least finds rest (verse 6). Verse 3 If a man begets a hundred children and lives many years, so that the days of his years are many, but his soul is not satisfied with goodness, or indeed he has no burial, I say that a stillborn child is better than he. The Jerusalem Bible (1966) and The Living Bible (1971) both suggest "as many daughters" as preferable translations in place of "lives many years". Retired Baptist minister Peter Pett suggests that, although Psalm 127 celebrates the blessing of many children, the number 100 should be interpreted as "a great many children, more than the norm". Insatiable longing (6:7–9) While people work to earn a living, Qoheleth argues that they are left unsatisfied because of the unanswered important questions: what is the point of wisdom, and what reason is there for the poor to find a role in life to improve one's lot? Verse 9 advises for contentment, because the desire to find answers beyond the obvious is seen as "vanity" (hebel). Verse 7 All the labour of man is for his mouth, and yet the appetite is not filled. "Appetite" is translated from , , which can also mean "soul, creature, living being, desire". The insatiability of desire is portrayed like "Sheol", the realm of death in the Hebrew Bible (cf. Isaiah 5:14; Habakkuk 2:5; Proverbs 27:20; 30:16). An impasse (6:10–12) This part starts a passage that ends in 7:13–14, insisting that 'no one knows what is good for humanity'.
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: where did they film hot tub time machine?
-A: Fernie Alpine Resort
-
-Q: who has the right of way in international waters?
-A: Neither vessel
-
-Q: who does annie work for attack on titan?
-A: Marley
-
-Q: when was the immigration reform and control act passed?
-A: November 6, 1986
-
-Q: when was puerto rico added to the usa?
-A: 1950
-
-Q: who had a baby at 100 in the bible?
-A:
-==================================================
-
-==================================================
-PROMPT #20:
-==================================================
-to the police authority in the applicant's district. Rifle and shotgun ownership permission can be given to "sober and responsible" persons 18 years or older. The applicant for the permission must document a need for the weapon. Three exceptions exist to this age qualification. Persons under the age of 18, but over 16 may apply for rifle or shotgun ownership licence with the consent of parents or guardian. For handguns, the ownership age is 21, but a license can be issued to those 18 or older if certain criteria are met. However, such license is valid for only 1 year and must be renewed to avoid revocation. For inherited weapons, it is up to the local police chief to make a decision based on the individual facts of the case. Obtaining a license There are two ways of obtaining an ownership license in Norway. The most common is through the process of obtaining a hunting license, the other is through a sports shooting license. For hunting To obtain a hunting license, the applicant must complete a 30-hour, 9-session course and pass a written multiple choice exam. The course includes firearm theory, firearm training, wildlife theory, and environmental protection training. Once the exam is passed, the applicant may enroll in the hunter registry and receive a hunting license. The membership must be renewed each year, through license payment. The hunting license is brought to the police station, where the applicant fills out an application for obtaining the proper firearm for his or her hunt. After evaluation, partmore than 90 days, or a fine of not more than $100.00, or both; Michigan Penal Code pg. 386 Sec. 223 (1). A person 18 years of age can legally purchase a muzzle loading pistol, cannon up to 42 lbs, or long gun from a private seller. A person 18 years of age can legally purchase a handgun from a private seller, but first must obtain a purchasing permit which is good for 30 days or have valid cpl license. Purchasing firearms At the age of 18 or up, it is legal to buy a handgun with a purchase license from a private seller, at the age of 21, it is legal to buy a firearm from a Federally licensed (FFL) dealer. No purchase license is required to purchase a long gun, or muzzle loader (a firearm that is more than 26 inches long) in Michigan. According to state law, a long gun may be purchased by anyone aged 18 or over who is not subject to restrictions based on criminal history, mental health history, or other disqualifying factor. A person must be at least 18 years old to purchase a long gun from a federal dealer or a private seller under Michigan law. Only Michigan residents may purchase handguns, sbr, sbs (meaning firearms that are 26 inches or less in length) in Michigan. Residents of any state may purchase long guns in Michigan and Michigan residents may purchase long guns in any state. NON-CPL Holders Only that are buying a handgun from a private seller:became C categorized. Category D2: No Storage restriction. Ownership and purchase conditions Category D requires the owner to be older than 18. Category C requires the owner to be older than 18, have hunting, shooting or ball-trap license of the current year stamped by a doctor. Category B requires the owner to be older than 18, be affiliated with a shooting range, have attended at least 3 shooting sessions with an instructor, and have a medical certificate. The shooter then receives a 5-year authorization to purchase and own of Category B firearms (and therefore Category C firearms since they are affiliated with a shooting range). Category A11 and A12 purchase not allowed since August 1st 2018, for gun ownership before its same requirement as Category B. No civilian may carry any weapons in a public place. A special form allows a civilian to apply for a 1-year carry license, which allows them to carry a handgun and a maximum of 50 rounds if they are "exposed to exceptional risks to their life". In practice, such authorizations are rare. Exceptions exist for children and teenagers with a shooting or ball-trap license and parental approval. A child aged between 9 and 12 can own D categorized weapon that shoot projectiles In a non pyrotechnic way between 2 and 20 joules. A teenager aged 12 to 16 can own C and D categorized weapons. They can also own one-shot, rimfire Category B firearms if they participate in international shooting competitions (only with a shooting licence). A shooter between the
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: where did they film hot tub time machine?
-A: Fernie Alpine Resort
-
-Q: who has the right of way in international waters?
-A: Neither vessel
-
-Q: who does annie work for attack on titan?
-A: Marley
-
-Q: when was the immigration reform and control act passed?
-A: November 6, 1986
-
-Q: when was puerto rico added to the usa?
-A: 1950
-
-Q: what age do you need to be to buy a bb gun?
-A:
-==================================================
-

benchmarks/data/prompts_g5/prompt_dump_trivia_hnsw.txt

@@ -1,484 +0,0 @@
-=== Prompt Dump for TRIVIA + HNSW ===
-Total prompts: 50
-Showing first 20 prompts:
-
-==================================================
-PROMPT #1:
-==================================================
-Jason Lee also portrays David Seville in live action/CGI films starring Alvin and the Chipmunks, which use a combination of live-action acting and computer animation. While Ross Bagdasarian Jr. does not do any voices for the film series, the films are all produced in association with Bagdasarian Productions, which owns the rights to all of the characters. Portrayed by Filmography Films Television See also References Fictional characters introduced in 1958 Alter egos Alvin and the Chipmunks Fictional managers Fictional producers American male characters in televisionRoss Dickran Bagdasarian (born May 6, 1949) is an American actor, animator and producer, known for his work on the Alvin and the Chipmunks franchise. He is the son of the franchise's creator, Ross Bagdasarian. Early life Bagdasarian was born in Fresno, California, the son of Armenian-American parents Armenuhi Bagdasarian (née Kulhanjian) and Ross Bagdasarian (1919–1972). As a child, he worked with his father on The Alvin Show by helping edit and coordinate the soundtracks and falsetto voice-overs of the Chipmunks. Career Bagdasarian graduated from law school. He succeeded his father as president of Bagdasarian Productions in 1972 after the death of the elder Bagdasarian. The company had fallen into obscurity after significant success between 1958 and the late 1960s. Bagdasarian was also admitted to the California bar as an attorney in 1975. Under Bagdasarian's supervision, new Chipmunks records were created shortly after his marriage to Karman, including Chipmunk Punk. In 1981, the Chipmunks returned to television in the cartoon special A Chipmunk Christmas. Two years later, Ruby-Spears Productions' Alvin and the Chipmunks Saturday morning cartoon series debuted on NBC. Based on that series, a feature film, The Chipmunk Adventure was released in 1987. Bagdasarian voices Alvin, Simon, and Dave Seville, and Karman voices Theodore and the Chipettes (Brittany, Jeanette, and Eleanor). Bagdasarian and Karman hold tight creative and financial control over the Chipmunk franchise, reviewing each and every business contract in great detail. In the mid-90s, Bagdasarian bought out his brother's and sister's portions of the Chipmunk rights, to take complete control of the franchise.Alvin and the Chipmunks, originally David Seville and the Chipmunks or simply The Chipmunks, are an American animated virtual band and media franchise first created by Ross Bagdasarian for novelty records in 1958. The group consists of three singing animated anthropomorphic chipmunks named Alvin, Simon, and Theodore who are originally managed by their human adoptive father, David "Dave" Seville. Bagdasarian provided the group's voices by producing sped-up recordings of his own, a technique pioneered on the successful "Witch Doctor". Later in 1958, Bagdasarian released the similarly-engineered "The Chipmunk Song" for which he came up with the chipmunk characters and their human father, attributing the track to them. David Seville and the Chipmunks released several more records over the following decade until Bagdasarian's death in 1972. The franchise was revived in 1979 with the characters' voices provided by his son Ross Bagdasarian Jr. and the latter's wife Janice Karman. Through the successful franchise, the Chipmunks have become one of the most successful children's artists of all time. It has garnered two number-one singles on the Billboard Hot 100 and won five Grammy Awards, having four Top 10 albums on the Billboard 200 and three certified platinum albums. "The Chipmunk Song" became one of the best-selling singles of all time at 5 million physical copies sold. The Chipmunks were first depicted in animated form in The Alvin Show (1961). The characters have since featured in several television series and films, as well as other media. In 2019, The Chipmunks received a star on the Hollywood Walk of Fame.
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Rita Coolidge sang the title song for which Bond film??
-A: Octopussy
-
-Q: Which Lloyd Webber musical premiered in the US on 10th December 1993??
-A: Sunset Boulevard
-
-Q: Who was the next British Prime Minister after Arthur Balfour??
-A: Campbell-Bannerman
-
-Q: Who had a 70s No 1 hit with Kiss You All Over??
-A: Exile
-
-Q: What claimed the life of singer Kathleen Ferrier??
-A: Cancer
-
-Q: Who was the man behind The Chipmunks?
-A: 
-==================================================
-
-==================================================
-PROMPT #2:
-==================================================
-and the drum set. Their film counterparts are Michelle and Eleni. Production history Broadway (2015-2019) Auditions began on January 19, 2015 for children ages nine through fifteen. Some recruiting was done through the School of Rock after-school educational program (which predated the film by several years) and open calls were held in New York at the Winter Garden, in Chicago and in Los Angeles. The production closed on January 20, 2019, after 1,309 performances. West End (2016–2020) On 7 December 2015, following the show's Broadway opening, it was announced by Andrew Lloyd Webber that the show would transfer to London's West End in autumn 2016, with the intention to open at the London Palladium. On 20 May 2016, the musical was confirmed at the Gillian Lynne Theatre instead of the Palladium with previews starting on 24 October 2016, opening night on 14 November 2016, and public booking opening on 25 May 2016. Lloyd Webber revealed that the production was able to open several months earlier than anticipated due to finding the child musician actors easily. Anna Louizos' scenery has been modified to fit the architecture of the Gillian Lynne Theatre from the traditional proscenium arch stage at Winter Garden Theatre. Changes include the removal of the pre-show curtain, the use of a revolving stage and action taking place in the aisles of the stalls. While the show remains to be set in America, the script has been adapted to include some minor references for a British audience. The original London cast includes David Fynn as DeweyThe Sound of Music, Camelot and Fiddler on the Roof played at the theatre in the early 1980s. In 1984, the interior was extensively modified by the introduction of a 'race track' that ran through the audience, for the show Starlight Express with performers on roller skates. The show premièred on 27 March, composed by Andrew Lloyd Webber and directed by Trevor Nunn and ran for 7,406 performances, over 18 years. With the removal of the 'tracks', the interior was extensively restored by architects Jaques Muir and Partners. This included the removal of 3,500 incandescent lamps that had become difficult to maintain and consumed a considerable amount of power. These were replaced by 88,000 low power LEDs specially designed for the theatre, creating the first auditorium completely lit in this way. Another Lloyd Webber production followed, Bombay Dreams premièred on 19 June 2002. It was created by A. R. Rahman with lyrics by Don Black and was directed by Steven Pimlott, closing after 1,500 performances on 13 June 2004. This was followed by the return to the West End of the Bee Gee's musical Saturday Night Fever on 6 July 2004, closing 22 October 2005 to tour. This was followed on 10 April 2006 by the jukebox musical Movin' Out, featuring the music of Billy Joel. This starred James Fox but ran for only two months. The Broadway musical Wicked received its London première at the venue on 27 September 2006 with a cast featuring Idina Menzel as Elphaba, Helen Dallimore as Glinda, Nigel Planer asand also starred comedian Tim Minchin as Judas Iscariot, former Spice Girl Melanie C as Mary Magdalene and BBC Radio 1 DJ Chris Moyles as King Herod. Tickets for most venues went on sale on 18 May 2012. In 2013, Lloyd Webber reunited with Christopher Hampton and Don Black on Stephen Ward the Musical. For his next project, a 2015 musical adaptation of the 2003 film School of Rock, auditions were held for children aged nine to fifteen in cooperation with the School of Rock music education program, which predated the film by several years. In April 2016, the English National Opera staged a revival of Sunset Boulevard at the London Coliseum. The limited run, semi-staged production directed by Lonny Price brought Glenn Close to reprise her star turn as "Norma Desmond", which was her first time performing the role in London; she had originated the role in Los Angeles in December 1993 and then on Broadway in November 1994 (which won her the 1995 Tony Award for Best Actress in a Musical). The 2016 London revival was so well-received that the production transferred to the Palace Theatre on Broadway in February 2017, making Lloyd Webber the first musical-theatre composer since 1953 to have four musicals running simultaneously on Broadway – a feat that his heroes Rodgers and Hammerstein had previously achieved. Lloyd Webber's memoir, Unmasked, was published in 2018. On 9 September 2018, Lloyd Webber, along with Tim Rice and John Legend each won an Emmy for Jesus Christ Superstar Live in Concert. With this
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Who was the man behind The Chipmunks??
-A: David Seville
-
-Q: Rita Coolidge sang the title song for which Bond film??
-A: Octopussy
-
-Q: Who was the next British Prime Minister after Arthur Balfour??
-A: Campbell-Bannerman
-
-Q: Who had a 70s No 1 hit with Kiss You All Over??
-A: Exile
-
-Q: What claimed the life of singer Kathleen Ferrier??
-A: Cancer
-
-Q: Which Lloyd Webber musical premiered in the US on 10th December 1993?
-A: 
-==================================================
-
-==================================================
-PROMPT #3:
-==================================================
-Cabinet Louis Botha, Prime Minister of the Union of South Africa (1910–1919) Behind Churchill are: George Barnes, leader of the National Democratic and Labour Party Sir Robert Borden, Prime Minister of Canada (1911–1920) To their right are: Arthur Balfour, 1st Earl of Balfour, former Prime Minister of the United Kingdom (1902–1905); First Lord of the Admiralty (1915–1916) and Foreign Secretary (1916–1919) (standing adlocutio in a black suit) H. H. Asquith, 1st Earl of Oxford and Asquith, Prime Minister of the United Kingdom (1908–1916) (sitting in front) Sir Eric Geddes, First Lord of the Admiralty (1917–1919) (behind, cleanshaven) Bonar Law, Leader of the Opposition (United Kingdom) (1911–1915), Secretary of State for the Colonies (1915–1916), Chancellor of the Exchequer (1916–1919) (later Prime Minister of the United Kingdom, 1922–1923) (dark moustache) Edward Morris, 1st Baron Morris, Prime Minister of Newfoundland (1909–1917) (white moustache, in the shadows) Herbert Kitchener, 1st Earl Kitchener, Secretary of State for War (1914–1916) (in the shadows) Bailey decided that the painting should include British and Dominion civilian leaders in office at the beginning and the end of the First World War. It includes Prime Ministers of Australia, Canada, Newfoundland, and New Zealand, and the Prime Ministers, Foreign Secretaries, Secretaries of War, and First Lords of the Admiralty of the United Kingdom, together with two leaders of the British Conservative and Labour parties. The Maharaja of Bikaner, a member of the Imperial War Cabinet and the Indian delegate to the Versailles Peace Conference, stands to the left next to Botha, both in military uniform. Kitchener standsArthur James Balfour, 1st Earl of Balfour, (, ; 25 July 184819 March 1930), also known as Lord Balfour, was a British Conservative statesman who served as Prime Minister of the United Kingdom from 1902 to 1905. As foreign secretary in the Lloyd George ministry, he issued the Balfour Declaration of 1917 on behalf of the cabinet, which supported a "home for the Jewish people" in Palestine. Entering Parliament in 1874, Balfour achieved prominence as Chief Secretary for Ireland, in which position he suppressed agrarian unrest whilst taking measures against absentee landlords. He opposed Irish Home Rule, saying there could be no half-way house between Ireland remaining within the United Kingdom or becoming independent. From 1891 he led the Conservative Party in the House of Commons, serving under his uncle, Lord Salisbury, whose government won large majorities in 1895 and 1900. An esteemed debater, he was bored by the mundane tasks of party management. In July 1902, he succeeded his uncle as prime minister. In domestic policy he passed the Land Purchase (Ireland) Act 1903, which bought out most of the Anglo-Irish land owners. The Education Act 1902 had a major long-term impact in modernising the school system in England and Wales and provided financial support for schools operated by the Church of England and by the Catholic Church. Nonconformists were outraged and mobilised their voters, but were unable to reverse it. In foreign and defence policy, he oversaw reform of British defence policy and supported Jackie Fisher's naval innovations. He secured the Entente Cordiale withthe county of Haddington. In October 1922 he, with most of the Conservative leadership, resigned with Lloyd George's government following the Carlton Club meeting, a Conservative back-bench revolt against continuance of the coalition. Bonar Law became prime minister. Like many Coalition leaders, he did not hold office in the Conservative governments of 1922–1924, but as an elder statesman, he was consulted by the King in the choice of Stanley Baldwin as Bonar Law's successor as Conservative leader in May 1923. His advice was strongly in favour of Baldwin, ostensibly due to Baldwin's being an MP but in reality motivated by his personal dislike of Curzon. Later that evening, he met a mutual friend who asked 'Will dear George be chosen?' to which he replied with 'feline Balfourian satisfaction,' 'No, dear George will not.' His hostess replied, 'Oh, I am so sorry to hear that. He will be terribly disappointed.' Balfour retorted, 'Oh, I don't know. After all, even if he has lost the hope of glory he still possesses the means of Grace.' Balfour was not initially included in Baldwin's second government in 1924, but in 1925, he returned to the Cabinet, in place of the late Lord Curzon as Lord President of the Council, until the government ended in 1929. With 28 years of government service, Balfour had one of the longest ministerial careers in modern British politics, second only to Winston Churchill . Last years Lord Balfour had generally good health until 1928 and remained until then a regular tennis player. Four years previously
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Who was the man behind The Chipmunks??
-A: David Seville
-
-Q: Which Lloyd Webber musical premiered in the US on 10th December 1993??
-A: Sunset Boulevard
-
-Q: Rita Coolidge sang the title song for which Bond film??
-A: Octopussy
-
-Q: Who had a 70s No 1 hit with Kiss You All Over??
-A: Exile
-
-Q: What claimed the life of singer Kathleen Ferrier??
-A: Cancer
-
-Q: Who was the next British Prime Minister after Arthur Balfour?
-A: 
-==================================================
-
-==================================================
-PROMPT #4:
-==================================================
-classic '70s pop song." In 1992, Mexican trio Pandora released a cover version titled "Pierdo el Control" on their album Ilegal. In 1979 Ginger Rogers sang this song on The Love Boat in the episode "Critical Success / The Love Lamp Is Lit / Take My Boyfriend, Please / Rent a Family / The Man in Her Life: Parts 1 & 2" In 2001, the film Get Over It featured a dance to this song at the beginning by some of the cast. References 1973 songs 1975 debut singles Songs written by Neil Sedaka Songs with lyrics by Howard Greenfield Neil Sedaka songs Captain & Tennille songs Andy Williams songs Number-one singles in Australia Billboard Hot 100 number-one singles Cashbox number-one singles RPM Top Singles number-one singles Grammy Award for Record of the Year A&M Records singles Juno Award for Best Selling Single singlesMusic Week rated the song four out of five, concluding, "A third huge hit for the boys." Tracklisting CD single "Kiss You All Over" (Radio Edit) - 4:31 "Kiss You All Over" (Club Mix) - 5:53 "Bonita" (Radio Edit) - 3:54 "Bonita" (Club Mix) - 7:08 Charts Release history References 1978 songs 1978 singles 1997 singles 1998 singles Billboard Hot 100 number-one singles Cashbox number-one singles Exile (American band) songs Number-one singles in New Zealand Number-one singles in South Africa Number-one singles in Australia Songs written by Mike Chapman Song recordings produced by Frank Farian Song recordings produced by Mike Chapman Songs written by Nicky Chinn RAK Records singles Curb Records singles Hilltak Records singles Warner Records singles Arista Records singles No Mercy (pop band) songs Songs about kissing Phyllis Hyman songs"Kiss You All Over" is a 1978 song performed by American group Exile, written by Mike Chapman and Nicky Chinn. It was included on the band's third album, Mixed Emotions (1978), and featured lead vocalist Jimmy Stokley and guitarist J.P. Pennington on vocals. On the American Top 40 broadcast of May 26, 1979, Casey Kasem reported that Chapman stated his source of inspiration for "Kiss You All Over" was "It's Ecstasy When You Lay Down Next to Me" by Barry White. The song was a number one single in the United States, but proved to be Exile's only big hit in the pop market (they would later have great success on the country music charts). It held the number one spot on the Billboard Hot 100 for four weeks (starting September 30), and Billboard ranked it as the No. 5 song for 1978. The track also reached number-one in at least three other nations. In the United Kingdom, the song was released on Mickie Most's RAK Records, and peaked at number 6 on the UK Singles Chart. The strings are played with a synthesizer in a backing track. In 2010, Billboard ranked the song tenth on its list of "The 50 Sexiest Songs of All Time". Lead vocalist on the number, Stokley was ousted from the band in 1979, his health declining thereafter until he died at the age of 41 in 1985. After the success of soft rock singles from the albums Mixed Emotions and All There Is, the band moved into country music in
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Who was the man behind The Chipmunks??
-A: David Seville
-
-Q: Which Lloyd Webber musical premiered in the US on 10th December 1993??
-A: Sunset Boulevard
-
-Q: Who was the next British Prime Minister after Arthur Balfour??
-A: Campbell-Bannerman
-
-Q: Rita Coolidge sang the title song for which Bond film??
-A: Octopussy
-
-Q: What claimed the life of singer Kathleen Ferrier??
-A: Cancer
-
-Q: Who had a 70s No 1 hit with Kiss You All Over?
-A: 
-==================================================
-
-==================================================
-PROMPT #5:
-==================================================
-21st century world: "We dislike low-lying voices, for one thing— contraltos now sound freakish and headmistressy, and even the majority of mezzo-sopranos should more accurately be categorised as almost-sopranos". However, she was "a singer of, and for, her time — a time of grief and weariness, national self-respect and a belief in human nobility". In this context "her artistry stands upright, austere, unfussy, fundamental and sincere". Shortly after Ferrier's death an appeal was launched by Barbirolli, Walter, Myra Hess and others, to establish a cancer research fund in Ferrier's name. Donations were received from all over the world. To publicise the fund a special concert was given at the Royal Festival Hall on 7 May 1954, at which Barbirolli and Walter shared the conducting duties without payment. Among the items was a rendition of Purcell's When I am laid in earth, which Ferrier had often sung; on this occasion the vocal part was played by a solo cor anglais. The Kathleen Ferrier Cancer Research Fund helped establish the Kathleen Ferrier Chair of Clinical Oncology at University College Hospital, in 1984. , it was continuing to fund oncology research. As the result of a separate appeal, augmented by the sales proceeds of a memoir edited by Neville Cardus, the Kathleen Ferrier Memorial Scholarship Fund was created to encourage young British and Commonwealth singers of either sex. The Fund, which has operated from 1956 under the auspices of the Royal Philharmonic Society, initially provided an annual award covering the cost of a year's study to a single prizewinner.In the course of her professional life the English contralto Kathleen Ferrier made a large number of recordings. In the summer of 1944 she signed a contract with Columbia, which lasted until February 1946. She then transferred to Decca, and remained with them until her death in October 1953. Apart from her studio recordings, many of her live performances and broadcast recitals were recorded, sometimes privately. Some of these were later issued as commercial recordings; others are held by individuals or in the archives of broadcasting companies. The following list is neither up to date nor entirely accurate, particularly in regard to a CD issue, entitled 'Kathleen Ferrier Remembered', released in June 2017, on SOMM264, comprising 26 tracks, 19 of which have never previously been issued. Most of these 19 are not listed below. They include Lieder by Schubert, Brahms, Wolf and Mahler and songs by Stanford, Parry, Jacobson and Rubbra, all taken from BBC broadcasts between 1947 and 1952. In April 2019, a recording of Ferrier singing in Bach's 'Magnificat' during the 1950 Vienna International Bach Festival was issued for the first time. The CD catalogue number is SOMM Ariadne 5004 and it also features Irmgard Seefried and Friedl Riegler (sopranos), Hugo Meyer-Welfing (tenor) and Otto Edelmann (bass). The Vienna Philharmonic Orchestra and Chorus of the Vienna State Opera are conducted by Volkmar Andreae. The existence of this recording was not known until a vinyl disc was offered for sale on an internet auction site in 2018. In superb recorded sound, this discovery is aKathleen Mary Ferrier, CBE (22 April 19128 October 1953) was an English contralto singer who achieved an international reputation as a stage, concert and recording artist, with a repertoire extending from folksong and popular ballads to the classical works of Bach, Brahms, Mahler and Elgar. Her death from cancer, at the height of her fame, was a shock to the musical world and particularly to the general public, which was kept in ignorance of the nature of her illness until after her death. The daughter of a Lancashire village schoolmaster, Ferrier showed early talent as a pianist, and won numerous amateur piano competitions while working as a telephonist with the General Post Office. She did not take up singing seriously until 1937, when after winning a prestigious singing competition at the Carlisle Festival she began to receive offers of professional engagements as a vocalist. Thereafter she took singing lessons, first with J.E. Hutchinson and later with Roy Henderson. After the outbreak of the Second World War Ferrier was recruited by the Council for the Encouragement of Music and the Arts (CEMA), and in the following years sang at concerts and recitals throughout the UK. In 1942 her career was boosted when she met the conductor Malcolm Sargent, who recommended her to the influential Ibbs and Tillett concert management agency. She became a regular performer at leading London and provincial venues, and made numerous BBC radio broadcasts. In 1946, Ferrier made her stage debut, in the Glyndebourne Festival premiere of Benjamin Britten's opera The Rape of Lucretia.
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Who was the man behind The Chipmunks??
-A: David Seville
-
-Q: Which Lloyd Webber musical premiered in the US on 10th December 1993??
-A: Sunset Boulevard
-
-Q: Who was the next British Prime Minister after Arthur Balfour??
-A: Campbell-Bannerman
-
-Q: Who had a 70s No 1 hit with Kiss You All Over??
-A: Exile
-
-Q: Rita Coolidge sang the title song for which Bond film??
-A: Octopussy
-
-Q: What claimed the life of singer Kathleen Ferrier?
-A: 
-==================================================
-
-==================================================
-PROMPT #6:
-==================================================
-"You Only Live Twice", performed by Nancy Sinatra, is the theme song to the 1967 James Bond film of the same name. The music was by veteran Bond film composer John Barry, with lyrics by Leslie Bricusse. The song is widely recognized for its striking opening bars, featuring a simple 2-bar theme in the high octaves of the violins and lush harmonies from French horns. It is considered by some to be among the best James Bond theme songs, and has become one of Nancy Sinatra's best known hits. Shortly after Barry's production, Sinatra's producer Lee Hazlewood released a more guitar-based single version. The song has been covered by many artists including Coldplay, Soft Cell, Björk and Shirley Bassey. In 1998, Robbie Williams re-recorded portions of the song (including the opening strings) for use in his UK number-one single "Millennium". Background James Bond veteran John Barry returned to the franchise to produce the score. The lyrics were by Leslie Bricusse, who had previously cowritten the lyrics for the theme to Goldfinger. An initial version of the song was performed by Julie Rogers and recorded with a 50 or 60 piece orchestra at CTS Studios. However, this version was not used since Barry decided to re-write and re-record the song: "It was usually the producers that said 'this isn't working, there's a certain something that it needed'. If that energy wasn't there, if that mysterioso kind of thing wasn't there, then it wasn't going to work for the movie." The Rogers song shares only two lines withBassey belting out the fantastic title song." He added that the remastered edition's sound quality was "impeccable". Chart positions Track listing Credits Project manager: Herb Agner Creative director: Michelle Azzopardi Composer, conductor, primary artist: John Barry Primary artist, vocals: Shirley Bassey Liner notes: Jeff Bond Composer, lyricist: Leslie Bricusse Project manager: Wendy Brueder Producer, reissue producer: Frank Collura Remastering: Bob Fisher Guitar, soloist: Vic Flick Art direction, design: Peter Grant Orchestra contractor: Sid Margo Lyricist: Anthony Newley A&R: Gregg Ogorzelec Engineer: John Richards Saxophone, soloist: John Scott Source: Aftermath Following the success of her performance on the title track, Shirley Bassey sang the title songs for two later Bond films, Diamonds Are Forever and Moonraker. John Barry used the Goldfinger theme on his 1965 John Barry Plays Goldfinger album that featured Robert Brownjohn artwork. References Footnotes Citations Bibliography Soundtrack albums from James Bond films Soundtrack 1964 soundtrack albums EMI Records soundtracks John Barry (composer) soundtracksJames Bond (Roger Moore), and the title evidently refers to the key aerial sequences featured in the movie. Prior to Rita Coolidge being assigned the Octopussy theme, Mari Wilson was a contender, a British singer whose retro-image evoked the mid-'60s when the Bond series originated; but Wilson's lack of a US-profile led to a negative decision. In January 1983, the producer of Octopussy: Cubby Broccoli, stated that he hoped to have current hitmaker Laura Branigan sing the movie's theme song, an artist choice which both Barry and Rice have stated would have pleased them. However, on March 29, 1983 Rita Coolidge was revealed as the singer, a seemingly surprising choice in that Coolidge's career peak had occurred some six years previously. Coolidge recalls that Barbara Broccoli, daughter of Cubby Broccoli and herself the assistant director of Octopussy, was a fan of Coolidge and made a point of playing Coolidge records around her father until "one day [he said], "Who is that? That's the voice I want for the movie." Rice still had to complete his contribution as the singer arrived in the studio, with Coolidge stating that "we were waiting for the lyrics as the instrumental track had already been done." The chorus of "All Time High" features a lyric similar to that of Coolidge's #2 hit "(Your Love Has Lifted Me) Higher and Higher" whose lyric "When you wrap your loving arms around me I can stand up and face the world again" is echoed by the "All Time High" lyric "We'll take on the
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Who was the man behind The Chipmunks??
-A: David Seville
-
-Q: Which Lloyd Webber musical premiered in the US on 10th December 1993??
-A: Sunset Boulevard
-
-Q: Who was the next British Prime Minister after Arthur Balfour??
-A: Campbell-Bannerman
-
-Q: Who had a 70s No 1 hit with Kiss You All Over??
-A: Exile
-
-Q: What claimed the life of singer Kathleen Ferrier??
-A: Cancer
-
-Q: Rita Coolidge sang the title song for which Bond film?
-A: 
-==================================================
-
-==================================================
-PROMPT #7:
-==================================================
-which allowed the first legal beer sales since the beginning of Prohibition on January 16, 1920. In 1933 state conventions ratified the Twenty-first Amendment, which repealed Prohibition. The Amendment was fully ratified on December 5, 1933. Federal laws enforcing Prohibition were then repealed. Dry counties Following repeal some states continued prohibition within their own jurisdictions. Almost two-thirds of the states adopted some form of local option which enabled residents in political subdivisions to vote for or against local prohibition. For a time, 38 percent of Americans lived in areas with Prohibition. By 1966, however, all states had repealed their statewide prohibition laws, with Mississippi the last state to do so. Notes Sources Walker, Robert S. and Samuel C. Patterson, Oklahoma Goes Wet: The Repeal of Prohibiton, Eagleton Institute, Rutgers University, (1961). External links Repeal Day is December Fifth See more related images by selecting the "Alcohol" subject at the Persuasive Cartography, The PJ Mode Collection, Cornell University Library Prohibition in the United States Economic history of the United States 1933 in the United States Articles containing video clipsimportation of alcoholic beverages in the United States. The resolution was sent to the states for ratification and became the Eighteenth Amendment to the U.S. Constitution. On January 8, 1918, Mississippi became the first state to ratify the amendment and on January 16, 1919, Nebraska became the 36th state to do so, securing its passage with the required three-fourths of the states. By the end of February 1919, only three states remained as hold-outs to ratification: New Jersey, Connecticut and Rhode Island. The National Prohibition Act, also known as the Volstead Act, was enacted on October 18, 1919. Prohibition in the United States went into effect on January 17, 1920. Nationwide prohibition was repealed in 1933 with the passage of the Twenty-first Amendment on February 20 and its ratification on December 5. List of formerly dry states This table lists the effective dates each state went dry and any dates of repeal that do not coincide with the end of national prohibition in 1933. See also Dry county Alcoholic beverage control state List of alcohol laws of the United States by state Notes Alcohol law in the United States Prohibition in the United StatesAugust 19. PPS functionals were completed August 21. GATV 5006 was then transferred to complex 14 for mating with the Atlas. July 27, 1966 (Wednesday) Following the announcement of his austerity programme, British Prime Minister Harold Wilson survived a vote of censure in the House of Commons, as members of his Labour Party (with an 88-seat majority) supported him. The final result was 246 votes in favor, and 325 against. On the same day, the nation's chief labor union, the Trades Union Congress, voted 20 to 12 in support of a resolution pledging to halt strikes that had been threatened during the six-month freeze against raising wages. For the first time in 58 years, liquor was legally served in Mississippi, the last of the United States to have repealed its prohibition laws. Effective July 1, individual local governments were allowed to hold referendum elections on whether to allow the sale of liquor at state-approved resorts, and Harrison County voters had endorsed the measure. At 6:55 p.m., after police cars escorted a liquor delivery truck into Biloxi. The first drink in the state was poured at the Broadwater Beach Hotel, and Louis Cobb, the first legal bartender in Mississippi, sold a glass of scotch whiskey to hotel manager T.M. Dorsett. Biloxi Mayor Dan Guice then cut the ribbon to open the entrance to the hotel's bar.Died: Brenda Sue Brown, 11, was beaten to death after walking with her sister to summer school in Shelby, North Carolina. Police were unable to charge a suspect with the crime, until
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Who was the man behind The Chipmunks??
-A: David Seville
-
-Q: Which Lloyd Webber musical premiered in the US on 10th December 1993??
-A: Sunset Boulevard
-
-Q: Who was the next British Prime Minister after Arthur Balfour??
-A: Campbell-Bannerman
-
-Q: Who had a 70s No 1 hit with Kiss You All Over??
-A: Exile
-
-Q: What claimed the life of singer Kathleen Ferrier??
-A: Cancer
-
-Q: What was the last US state to reintroduce alcohol after prohibition?
-A: 
-==================================================
-
-==================================================
-PROMPT #8:
-==================================================
-to New York City for work in summer stock theatre shortly before winning a supporting role in MGM's These Glamour Girls (1939) opposite Lana Turner and Lew Ayres. The role of Betty was said to have been written especially with Hunt in mind. Other roles in major studio productions soon followed, including supporting roles as Mary Bennet in MGM's version of Pride and Prejudice (1940) with Laurence Olivier, and as Martha Scott's surrogate child Hope Thompson in Cheers for Miss Bishop (1941). Years at MGM In 1941, Hunt signed a contract with MGM, where she remained for the next six years. While filming Blossoms in the Dust, film director Mervyn LeRoy lauded Hunt for her heartfelt and genuine acting ability. During this period she had starring roles in 21 films, including The Penalty (1941) opposite Lionel Barrymore, Panama Hattie (1942) opposite Ann Sothern and Red Skelton, and the war drama Pilot No. 5 (1943) in which she was cast as the love interest of Franchot Tone, and The Valley of Decision (1945). In 1944 she polled seventh in a list by exhibitors of "Stars of Tomorrow". She previously did a screen test to play Melanie Hamilton in Gone with the Wind (1939) and was told by David O. Selznick she would play the role, but to "keep it a secret for now." Three days later, it was announced that Olivia de Havilland was cast. In 1944, she appeared in None Shall Escape, a film that is now regarded as the first about the Holocaust. She playedMiss America 1941, the 15th Miss America pageant, was held at the Boardwalk Hall in Atlantic City, New Jersey on September 6, 1941. Shortly after the crowning of Miss California, Rosemary LaPlanche, who had been first runner-up in 1940, the pageant committee adopted this rule: "No contestant can compete in Atlantic City for the title of Miss America more than once", thus eliminating future state winners with more than one attempt at the national title. LaPlanche became a film actress, as did her sister, Louise LaPlanche. 1941 was also the first year that the special award, “Miss Congeniality” was created. It went to Mifaunwy Shunatona, a member of the Otoe and Pawnee tribes — she was also the first American Indian contestant in the pageant’s history. Results Awards Preliminary awards Other awards Contestants References Secondary sources External links Miss America official website 1941 1941 in the United States 1941 in New Jersey September 1941 events Events in Atlantic City, New JerseyMiss America 1942, the 16th Miss America pageant, was held at the Warner Theater in Atlantic City, New Jersey on September 12, 1942. Miss Texas, Jo-Carroll Dennison won the title after winning the swimsuit and talent categories. She was the first Miss Texas to win the Miss America title. Dennison became an actress and had roles in films such as Winged Victory. She was married at one time to comedian Phil Silvers. Results Awards Preliminary awards Other awards Contestants References Secondary sources External links Miss America (1942) 1942 1942 in the United States 1942 in New Jersey September 1942 events Events in Atlantic City, New Jersey
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Who was the man behind The Chipmunks??
-A: David Seville
-
-Q: Which Lloyd Webber musical premiered in the US on 10th December 1993??
-A: Sunset Boulevard
-
-Q: Who was the next British Prime Minister after Arthur Balfour??
-A: Campbell-Bannerman
-
-Q: Who had a 70s No 1 hit with Kiss You All Over??
-A: Exile
-
-Q: What claimed the life of singer Kathleen Ferrier??
-A: Cancer
-
-Q: Which actress was voted Miss Greenwich Village in 1942?
-A: 
-==================================================
-
-==================================================
-PROMPT #9:
-==================================================
-De Tokyo Stock Price Index (Japans: 東証株価指数) of TOPIX is een belangrijke aandelenindex van de Tokyo Stock Exchange. Berekening In deze index zijn alle bedrijven opgenomen die op de beurs van Tokio staan genoteerd in de First Section. Dit zijn de grootste en meest liquide aandelen die op de beurs worden verhandeld. Tot medio 2006 werd het gewicht van de individuele bedrijven in de index bepaald op basis van de marktkapitalisatie, hierna wordt ook de free float in de berekening meegenomen. Het effect van deze verandering was significant, daar veel Japanse bedrijven aandelen houden in andere Japanse bedrijven, ook wel bekend als crossholdings, om daarmee de langdurige zakenrelatie te onderstrepen. Deze belangen worden voor lange tijd gehouden en worden niet tot de free float gerekend. De index heeft 4 januari 1968 als startdatum, maar ging op 1 juli 1969 daadwerkelijk van start. Een andere belangrijke beursindex in Japan is de Nikkei 225. In deze index zijn 225 bedrijven opgenomen en dit is een prijsgewogen index. Samenstelling Eind maart 2021 bestond de index uit 2187 aandelen. Door het grote aantal aandelen is het gewicht van de individuele namen zeer klein. De top 10 aandelen hebben een gezamenlijk gewicht in de index van slechts 18,4% en de lijst zag er als volgt uit, met de gewichten tussen de haakjes: De belangrijkste drie sectoren zijn: elektronische apparatuur, informatie technologie en chemie. Deze drie vertegenwoordigen tezamen zo'n 34% van de index, waarvan de sector elektronische apparatuur het grootst is met een gewicht van 17,5%. Koershistorie De hoogste stand van deTOPIX steht für Tōkyō Stock Price Index (jap. , Tōshō kabuka shisū) und ist neben dem Nikkei 225 ein Kursindex der Tokioter Börse. Berechnet wird der TOPIX seit dem 1. Juli 1969. Die Index-Basis liegt bei 100 Punkten per 4. Januar 1968. Er enthält alle japanischen Aktien, welche im amtlichen Handel zugelassen sind. Die Gewichtung der einzelnen Unternehmen im Index erfolgt anhand der Marktkapitalisierung. Gegenwärtig (8. September 2021) setzt sich der Index aus 2.189 Aktien zusammen. Wegen dieser hohen Zahl an vertretenen Unternehmen wird der TOPIX als aussagekräftiger für den Zustand der japanischen Wirtschaft angesehen als der Nikkei 225. Weblinks Beschreibung des TOPIX (engl.) TOPIX in Echtzeit Jährliche Entwicklung des TOPIX seit 1949 (Daten vor 1969 – dem Einführungsjahr des TOPIX – sind rückgerechnet; XLS-Format, 31,5 KB; abgerufen am 12. Oktober 2017) Einzelnachweise Aktienindex Wirtschaft (Japan) Abkürzung, commonly known as TOPIX, along with the Nikkei 225, is an important stock market index for the Tokyo Stock Exchange (TSE) in Japan, tracking all domestic companies of the exchange's Prime market division. It is calculated and published by the TSE. , there were 1,669 companies listed on the First Section of the TSE, and the market value for the index was ¥197.4 trillion. The index transitioned from a system where a company's weighting is based on the total number of shares outstanding to a weighting based on the number of shares available for trading (called the free float). This transition took place in three phases starting in October 2005 and was completed in June 2006. Although the change is a technicality, it had a significant effect on the weighting of many companies in the index, because many companies in Japan hold a significant number of shares of their business partners as a part of intricate business alliances, and such shares are no longer included in calculating the weight of companies in the index. The TOPIX index is traded as a future on the Osaka Exchange under the ticker symbol JTPX. The CQG contract specifications for the TOPIX Index are listed below. TSE currently calculates and distributes TOPIX every second and further plans to launch a new High-Speed Index dissemination service provided at the millisecond level starting from February 28, 2011. History of TOPIX 1969-07-01 TSE to begin calculating and publishing “TOPIX” and “TOPIX Sector Indices” 1969-08-18 TSE to begin calculating and publishing “Tokyo Stock
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Who was the man behind The Chipmunks??
-A: David Seville
-
-Q: Which Lloyd Webber musical premiered in the US on 10th December 1993??
-A: Sunset Boulevard
-
-Q: Who was the next British Prime Minister after Arthur Balfour??
-A: Campbell-Bannerman
-
-Q: Who had a 70s No 1 hit with Kiss You All Over??
-A: Exile
-
-Q: What claimed the life of singer Kathleen Ferrier??
-A: Cancer
-
-Q: What is the Japanese share index called?
-A: 
-==================================================
-
-==================================================
-PROMPT #10:
-==================================================
-Man in the Music: The Creative Life and Work of Michael Jackson is a non-fiction book written by Joseph Vogel, published in June 2011 by the Sterling Publishing. Reception Man in the Music: The Creative Life and Work of Michael Jackson, was described by the Associated Press as "a fascinating read and really a must have for any fan of Jackson." Filmmaker Spike Lee characterized it as having "brilliantly cracked the DNA, the code, the artistry of Michael Joseph Jackson." References Works about Michael Jackson 2011 non-fiction books Sterling Publishing booksMoonwalk is a 1988 autobiography written by American recording artist Michael Jackson. The book was first published by Doubleday on February 1, 1988, five months after the release of Jackson's 1987 Bad album, and named after Jackson's signature dance move, the moonwalk. The book contains a foreword by Jacqueline Onassis. It reached number one on the New York Times Best Seller list. The book was reissued by Doubleday on October 13, 2009, following Jackson's death on June 25, 2009. Production Jacqueline Onassis, who was an editor at Doubleday, secured the book deal and paid Jackson a $300,000 advance. As part of the deal Jackson wanted Onassis to write a foreword, which she initially refused not wanting her name on any books she worked on but agreed to three paragraphs. She also edited the book. The first manuscript of the book was written by Robert Hilburn and was refused by the publishers, Doubleday, because it lacked "juicy details". A second manuscript was written by Stephen Davis, which Jackson drastically edited. Jackson finally decided to write the book himself, with help from Shaye Areheart. Due to the public interest in Jackson, Moonwalk was prepared for publication in secret. Relatives of Doubleday employees were hired as couriers, to deliver portions of the book from the company's head office in Manhattan to the printing plant in Fairfield, Pennsylvania. At the printing plant, the book was given the code name "Neil Armstrong", after the first "moonwalker". Narrative Dedicated to Fred Astaire, the book discusses Jackson's show business friends, girlfriends and hisMichael Jackson: Unauthorized in a 1994 biography of the late pop star Michael Jackson, written by celebrity biographer Christopher Andersen. Development According to Andersen, work started on the book in early 1991 when he received a call from a fellow journalist, who told him that two workers at Jackson's Neverland Ranch allegedly witnessed Jackson fondling a young celebrity. Andersen tried to interview Jackson several times, but was turned down. When Michael was publicly accused of child molestation in 1993, Andersen was told that he was under surveillance from investigators. Reception The book was largely overlooked by the public. Dana Kennedy of Entertainment Weekly felt that, with its "killer material", Anderson "probably could have retired from the celebrity-bio grind for good" had it been released five years before. People magazine found it to be a "sad book", considering its dark revelations about Jackson's behaviour. References 1994 non-fiction books Unauthorized biographies Works about the Michael Jackson sexual abuse allegations Biographies about musicians
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Who was the man behind The Chipmunks??
-A: David Seville
-
-Q: Which Lloyd Webber musical premiered in the US on 10th December 1993??
-A: Sunset Boulevard
-
-Q: Who was the next British Prime Minister after Arthur Balfour??
-A: Campbell-Bannerman
-
-Q: Who had a 70s No 1 hit with Kiss You All Over??
-A: Exile
-
-Q: What claimed the life of singer Kathleen Ferrier??
-A: Cancer
-
-Q: What was the name of Michael Jackson's autobiography written in 1988?
-A: 
-==================================================
-
-==================================================
-PROMPT #11:
-==================================================
-including popular titles by Sérgio Mendes and Herb Alpert were released with this audio process starting in September 1968. Other record labels soon followed suit, and an estimated 10% of all stereophonic albums released during the late 1960s and early 1970s employed the system. Other labels known to have used the system include Warner Bros. Records and Reprise Records. One of the biggest selling albums using the process is The Association's Greatest Hits, released in 1968. This recording has sold more than 2 million copies in the United States. The process was also used on the 1968 Frank Sinatra album Cycles as well as on most of the studio recordings on Wheels of Fire by Cream. Early 1968 copies of Neil Young's self-titled debut album also used the system. Use of Haeco-CSG in promotional recordings for radio The original intention of using Haeco-CSG on commercial LP releases was rather short lived, however, use of the process continued well into the mid-1970s on promotional records sent to radio stations. Many commercial FM Rock stations did not transition from mono to stereo broadcasting until the mid to late 1970s. AM Pop music stations continued to broadcast in mono, as AM stereo broadcasting was not introduced until 1982 and was never widely adopted. Many promotional singles and some commercial singles from the Warner/Reprise/Atlantic label group from this era had "CSG Mono Process" or "CSG Process" printed on the labels. Artists included Frank Sinatra, Gordon Lightfoot, James Taylor, Seals and Crofts. Warner subsidiary labels such as Atlantic issued a serieswas introduced to the public on December 13, 1957, at the Times Auditorium in New York City. 500 copies of this initial demonstration record were pressed. On December 16, 1957, Frey advertised in the trade magazine Billboard that he would send a free copy to anyone in the industry who wrote to him on company letterhead. Frey became known as "Mr. Stereo" during that era. Stereophonic sound was not entirely new to the public. In 1952 sound engineer Emory Cook developed a "Binaural" disk that used two separate grooves and playback needles to produce stereophonic sound; the following year he had a catalog of about 25 disks available for audiophiles. Multi-channel sound was integral to the widescreen motion picture processes Cinerama (1952) and CinemaScope (1953). Stereophonic audio tapes had been commercially available to audiophiles, although expensive, since the early-1950s. After the release of the Audio Fidelity demonstration disks, the other spur to the popularity of stereo disks was the reduction in price of a stereo magnetic cartridge, for playing the disks, from $250 to $29.95 in June 1958. The first four stereo discs available to the general public were released by Audio Fidelity in March, 1958--Johnny Puleo and his Harmonica Gang Volume 1 (AFSD 5830), Railroad - Sounds of a Vanishing Era (AFSD 5843), Lionel - Lionel Hampton and his Orchestra (AFSD 5849) and Marching Along with the Dukes of Dixieland Volume 3 (AFSD 5851). By the end of March the company had four more stereo LPs available. In the summer of 1958, Audio Fidelity recordedin 1957, with his Essex Records office manager George Phillips, he founded Somerset Records and Somerset Stereo Fidelity Records budget albums. His greatest claim to fame was selling large amounts of cheaply priced albums, with Somerset claiming to have manufactured the first stereo budget albums. The name of Somerset high fidelity albums was suggested by Miller International's West Coast distributor, Jimmy Warren, with the name of Stereo Fidelity (stereo albums) thought of by Wally Hill to capitalize on the public's interest in both high fidelity and stereophonic sound. The economy came from Miller starting his own record factory in Swarthmore, Pennsylvania, using public domain music and non union musicians from outside the United States to record cover versions of hit songs of the time. Many original tunes were written by Monty Kelly, Robert Lowden, and Joseph Kuhn with the music published by Miller's own music publisher, Chesdel Music created in 1962. Miller had his own distribution channels of his records in supermarkets and drugstores with the cheap albums being sold in metal racks similar to those holding paperback books or cardboard record holders called "dumps" that could be placed anywhere. Miller's record albums were sold wholesale for 93 cents to salesmen who sold them to merchants who sold them to the public for $1.98. Somerset Records used artist Anthony "Chic" Laganella to create attractive eye catching album covers. Miller used the name 101 Strings for several German orchestras; their first album appearing in September 1957. In 1958 Somerset released 24 101 Strings titles. Miller International's philosophy
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Who was the man behind The Chipmunks??
-A: David Seville
-
-Q: Which Lloyd Webber musical premiered in the US on 10th December 1993??
-A: Sunset Boulevard
-
-Q: Who was the next British Prime Minister after Arthur Balfour??
-A: Campbell-Bannerman
-
-Q: Who had a 70s No 1 hit with Kiss You All Over??
-A: Exile
-
-Q: What claimed the life of singer Kathleen Ferrier??
-A: Cancer
-
-Q: In which decade did stereo records first go on sale?
-A: 
-==================================================
-
-==================================================
-PROMPT #12:
-==================================================
-Flack in 1896) to win gold medals in both the 800 m and 1500 m in the same Olympics. Billy Mills, an unfancied runner, became the only American to win the gold in the men's 10,000 m. Bob Hayes won the 100 metre title in a time of 10.06 seconds, equaling the world record, and set the current record for the fastest relay leg in the 4×100 m. Joe Frazier, future heavyweight champion of the world, won a gold medal in heavyweight boxing while competing with a broken thumb. This was the last Summer Olympics to use a cinder running track for athletic events, and the first to use fiberglass poles for pole vaulting. Zambia declared its independence on the day of the closing ceremony of the 1964 Summer Olympics, thereby becoming the first country ever to have entered an Olympic games as one country, and left it as another. This was celebrated in the ceremony itself by the team using a placard with "Zambia" instead of the "Northern Rhodesia" placard from the opening ceremony. Zambia was the only team to use a placard in the closing ceremony. The start of operations for the first Japanese "bullet train" (the Tōkaidō Shinkansen) between Tokyo Station and Shin-Ōsaka Station was scheduled to coincide with the Olympic games. The first regularly scheduled train ran on 1 October 1964, just nine days before the opening of the games, transporting passengers in about four hours, and connecting the three major metropolitan areas of Tokyo, Nagoya, and Osaka. Ranatunge Karunananda who representedsystems were used: official hand timing, hand started photo-finish times, and the Gustavus Town Kirby timing device, which was designed by Kirby to determine the correct order of finish in horse races. The official report for 1932 Olympics states: "In addition to hand timing, two auxiliary electrical timing devices were used. Both were started by an attachment to the starters gun. One was stopped by hand at the time the runners hit the tape. The other was provided with a motion picture camera which photographed the runner at the tape and the dial of the time indicator simultaneously." Kirby's system was also used at the 1932 US. Olympic Trials, where Ralph Metcalfe's winning time of 10.62 in the 100 meters is considered possibly the first automatically timed world record. FAT was also used in 1936, but very few times have been found. In 1948, Bulova began developing the Phototimer, a unique combination of photo-finish camera and precision electronic timing instrument. The Phototimer was the first automatic timing device to be used in competitive sports. It was used extensively in North America, including at the 1948 US Olympic trials. The Bulova device was activated by the sound of the starting gun firing, rather than by a direct connection, which means that the times were around 0.02 seconds faster than reality. The 1948 Olympics, however, continued to use Omega timing with a device called the 'Magic Eye', developed by British Race Finish Recording Co. Ltd. The automatic times produced in the 1948 Olympics have never been released, butWhile the most notable story coming out of 1968 was socio-political, politics involved with the Olympics was not something unique to this year. However, the year marked the beginning of several emerging elements of contemporary track and field. Automatic timing While timing to the 100th of a second had been experimented with for many years, the 1968 Summer Olympics were the first to use Fully Automatic Timing, in not only athletics, but in canoeing, rowing, cycling, equestrian and swimming competitions. Subsequently, systems to record such times became more common and thus the accuracy of Fully Automatic Timing became mandated for World Record acceptance. While this rule was officially put into place in 1977, many 1968 records still stood as the first Automatically timed record. All weather tracks This technology too had been developing, but Tartan tracks were used as the competition surface for the first time at an Olympics. Since then an all-weather running track was required for all top-level competition. Subsequently, the inconsistency of the running surface became a significantly smaller factor in athletic performance. Altitude With the Olympics happening in Mexico City, at high altitude, the effect of the thin air on athletic performance became a factor on world records. This was already a known phenomenon, and the American team was selected by holding the Olympic Trials at high altitude at Echo Summit, California. In 1955, Lou Jones set the world record in the 400 meters at altitude in Mexico City. Following the 1968 Summer Olympics the: Men's 100 meters record, set by Jim
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Who was the man behind The Chipmunks??
-A: David Seville
-
-Q: Which Lloyd Webber musical premiered in the US on 10th December 1993??
-A: Sunset Boulevard
-
-Q: Who was the next British Prime Minister after Arthur Balfour??
-A: Campbell-Bannerman
-
-Q: Who had a 70s No 1 hit with Kiss You All Over??
-A: Exile
-
-Q: What claimed the life of singer Kathleen Ferrier??
-A: Cancer
-
-Q: In what year's Olympics were electric timing devices and a public-address system used for the first time?
-A: 
-==================================================
-
-==================================================
-PROMPT #13:
-==================================================
-A list of stratovolcanoes follows below. Africa Cameroon Mount Cameroon Democratic Republic of Congo Mount Nyiragongo, Goma; designated as a Decade Volcano It contains an active lava lake inside its crater which overflowed due to cracks in 2002. Mount Mikeno Eritrea Alid Volcano Dubbi Volcano Nabro Volcano Ethiopia Adwa Borawli, Afar Region Dabbahu Volcano Mount Fentale Kenya Mount Kenya, which contains several volcanic plugs on its peak. Mount Longonot Rwanda Mount Bisoke, on the border between Rwanda and the Democratic Republic of the Congo. Mount Gahinga, on the border between Rwanda and Uganda. Mount Karisimbi, on the border between Rwanda and the Democratic Republic of the Congo. Mount Muhabura, on the border between Rwanda and Uganda. Mount Sabyinyo, marks the border between Rwanda, Uganda, and the Democratic Republic of the Congo. Tanzania Ol Doinyo Lengai, the Earth's only active carbonatite lava-producing volcano. Mount Kilimanjaro, a dormant stratovolcano. It is the highest point of Africa. Mount Meru Mid-Atlantic Ridge Mount Pico in Pico Island, Azores, Portugal Teide in Tenerife, Canary Islands, Spain; designated as a Decade Volcano Cumbre Vieja in La Palma, Canary Islands, Spain Mount Fogo in Fogo, Cape Verde Green Mountain, Ascension Island Pico de las Nieves in Gran Canaria, Canary Islands, Spain Americas Caribbean La Grande Soufrière on Basse-Terre Island, Guadeloupe Soufriere Hills on the island Montserrat Its 1995 eruptions resulted in the abandonment of its capital city, Plymouth. Soufrière on the island Saint Vincent Mount Pelée on the island Martinique Its devastating eruption on 8 May 1902 resulted in the complete destruction ofMount Kilimanjaro is a volcano in Tanzania and the highest mountain in Africa. Kilimanjaro may also refer to: Tanzania Kilimanjaro National Park comprises the whole of Mount Kilimanjaro above the tree line and six forest corridors stretching down Kilimanjaro Region, a region in Tanzania Kilimanjaro (ward), a ward in the Moshi Urban district of Kilimanjaro Region, Tanzania Kilimanjaro International Airport in Tanzania a Tanzanian beer, see Beer in Africa#Eastern Africa a Tanzanite jewellery brand owned by F. Hinds Music Killamanjaro, a Jamaican reggae sound system Albums Kilimanjaro, an album by German artist Superpitcher Kilimanjaro (The Rippingtons album), a 1988 album by The Rippingtons Kilimanjaro (The Teardrop Explodes album), an album by The Teardrop Explodes Songs "Kilimanjaro", song by The Del Vikings 1962 "Kilimanjaro", song by Manhattan Brothers 1955 "Kilimanjaro", song by The Teardrop Explodes 1980 "Kilimanjaro", song by Juluka 1984 "Kilimandjaro" (song), a 1966 French-language song by French singer Pascal Danel "Kilimanjaro" (song), a 2010 song by A.R. Rahman from the film Enthiran "Kilimanjaro", a song by KSI from the 2016 extended play Keep Up Film Kilimanjaro (film), a 2013 American film Nigeria Kilimanjaro restaurant, a fast-food chain in Nigeria. See also The Snows of Kilimanjaro (disambiguation)Mount Kilimanjaro () is a dormant volcano located in Kilimanjaro Region of Tanzania. It has three volcanic cones: Kibo, Mawenzi, and Shira. It is the highest mountain in Africa and the highest single free-standing mountain above sea level in the world: above sea level and about above its plateau base. It is the highest volcano in Africa and the Eastern Hemisphere. Kilimanjaro is the fourth most topographically prominent peak on Earth. It is part of Kilimanjaro National Park and is a major hiking and climbing destination. Because of its shrinking glaciers and ice fields, which are projected to disappear between 2025 and 2035, it has been the subject of many scientific studies. Toponymy The origin of the name Kilimanjaro is not known, but a number of theories exist. European explorers had adopted the name by 1860 and reported that Kilimanjaro was the mountain's Kiswahili name. The 1907 edition of The Nuttall Encyclopædia also records the name of the mountain as Kilima-Njaro. Johann Ludwig Krapf wrote in 1860 that Swahilis along the coast called the mountain Kilimanjaro. Although he did not offer any support, he claimed that Kilimanjaro meant either mountain of greatness or mountain of caravans. Under the latter meaning, kilima meant mountain and jaro meant caravans. Jim Thompson claimed in 1885, again without support, that the term Kilima-Njaro "has generally been understood to mean" the mountain (kilima) of greatness (njaro). He also suggested "though not improbably it may mean" the white mountain. Njaro is an ancient Kiswahili word for shining. Similarly, Krapf wrote that a
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Who was the man behind The Chipmunks??
-A: David Seville
-
-Q: Which Lloyd Webber musical premiered in the US on 10th December 1993??
-A: Sunset Boulevard
-
-Q: Who was the next British Prime Minister after Arthur Balfour??
-A: Campbell-Bannerman
-
-Q: Who had a 70s No 1 hit with Kiss You All Over??
-A: Exile
-
-Q: What claimed the life of singer Kathleen Ferrier??
-A: Cancer
-
-Q: Which volcano in Tanzania is the highest mountain in Africa?
-A: 
-==================================================
-
-==================================================
-PROMPT #14:
-==================================================
-of the Libyan Draft Constitutional Charter for the Transitional Stage: The national flag shall have the following shape and dimensions: Its length shall be double its width, its shall be divided into three parallel coloured stripes, the uppermost being red, the centre black and lowest green, the black stripe shall be equal in area to the other two stripes together and shall bear in its centre a white crescent, between the two extremities of which there shall be a five-pointed white star. On 10 March 2011, France was the first country to recognise the council as the official government of Libya, as well as the first to allow the Libyan embassy staff to raise the flag. On 21 March, the flag was flown by the Permanent Mission of Libya to the United Nations and appeared on their official website, and thereafter in late August by the Arab League and by Libya's own telecommunications authority, the Libya Telecom & Technology, on its own website. In the following months many other Libyan embassies replaced the green flag of Gaddafi with the tricolour flag. This original flag of Libya is now the only flag used by the United Nations to represent Libya, according to the following UN statement: "Following the adoption by the General Assembly of resolution 66/1, the Permanent Mission of Libya to the United Nations formally notified the United Nations of a Declaration by the National Transitional Council of 3 August 2011 changing the official name of the Libyan Arab Jamahiriya to 'Libya' as well as athe flag's colours and symbols. According to Omar Faiek Shennib, "red was selected for the blood sacrificed for the freedom of Libya, black to remember the dark days that Libyans lived under the occupation of the Italians and green to represent its primary wealth, agriculture, [Libya once being referred to as the 'agricultural basket' or 'breadbasket' of the Ottoman Empire] and the future prosperity of the country. The star and crescent were placed within the black central strip of the flag as a reference to the Senussi flag and the role of King Idris in leading the country to independence". The flag's colours also echo the colours of the flags of the three regions of Libya: Fezzan (red), Cyrenaica (black), and Tripolitania (green). Under Muammar Gaddafi's dictatorship, Libya had a red-white-black flag from 1969 to 1977, and it was replaced by the all-green flag from 1977 to 2011, during which it was the only flag in the world to have one color and no design. During the Libyan Civil War against the rule of Muammar Gaddafi, the 1951–69 flag – as well as various makeshift versions without the crescent and star symbol, or without the green stripe – came back into use in areas held by the Libyan opposition and by protesters at several Libyan diplomatic missions abroad. The National Transitional Council, formed on 27 February 2011, adopted the flag previously used in the Kingdom of Libya between 1951 and 1969 as the "emblem of the Libyan Republic". The flag was officially defined in article threeThe flag of Libya from 1977 to 2011 was used by the Socialist People's Libyan Arab Jamahiriya from 1977 to 1986 and later the Great Socialist People's Libyan Arab Jamahiriya until 2011. The design is a green field in 1:2 ratio and was considered the only solid colour national flag in the world during its time. In 2011, after the collapse of Gaddafi's government, the 1951–1969 flag from the Kingdom of Libya was re-adopted but the flag introduced by Gaddafi remained in use by Pro-Gaddafists and Gaddafi loyalists. Before 1977, the country was called the Libyan Arab Republic from 1969 to 1977 and used a red-white-black flag similar to most traditional Arab national flags bearing a resemblance to the modern flag of Yemen. in 1977 after the Egyptian-Libyan War, the blank green flag was introduced to replace the red-white-black flag to avoid similarities with Egypt. History of Libya under Muammar Gaddafi Flags introduced in 1977 1977 establishments in Libya 2011 disestablishments in Libya
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Who was the man behind The Chipmunks??
-A: David Seville
-
-Q: Which Lloyd Webber musical premiered in the US on 10th December 1993??
-A: Sunset Boulevard
-
-Q: Who was the next British Prime Minister after Arthur Balfour??
-A: Campbell-Bannerman
-
-Q: Who had a 70s No 1 hit with Kiss You All Over??
-A: Exile
-
-Q: What claimed the life of singer Kathleen Ferrier??
-A: Cancer
-
-Q: The flag of Libya is a plain rectangle of which color?
-A: 
-==================================================
-
-==================================================
-PROMPT #15:
-==================================================
-la Francophonie. Places of worship Niger being a predominantly Muslim country, mosques are the most common places of worship, with the Grande Mosquée being the largest in the city. There are also various Christian churches, most notably Our Lady of Perpetual Help Cathedral and the Cathedral de Maourey. Governance Administration Niamey makes up a special capital district of Niger, which is surrounded by the Region of Tillabéri. The city of Niamey itself is governed as an autonomous first-level administrative block, the Niamey Urban Community (Fr. Communauté Urbaine de Niamey, or CUN). It includes five Urban Communes, divided into 44 "Districts" and 99 "Quartiers", including formerly independent towns. It is a co-equal first division subdivision with the seven Regions of Niger. The Niamey Urban Community includes an administration and Governor appointed by national leaders. Like the rest of Niger, Niamey has seen a decentralisation of governance since 2000. Government Ordinance n°2010–56 and Presidential Decree n°2010-679 of September 2010 mandated an elected City Council for the city of Niamey, subsumed under the CUN. This excludes some outlying areas of the CUN. Forty-five councillors are popularly elected and in turn elect the Mayor of the City of Niamey. In July 2011, the first Mayor under the new system, Oumarou Dogari Moumouni, was installed by the Governor of the CUN Mrs. Aïchatou Boulama Kané and the City Council. The City Council and Mayor have limited roles compared to the CUN Governor. Niamey has a third layer of government in the Commune system. Each Commune elects its own council, and outsidein Niger Niamey NigerNiamey () is the capital and largest city of Niger. Niamey lies on the Niger River, primarily situated on the east bank. Niamey's population was counted as 1,026,848 as of the 2012 census. As of 2017, population projections show the capital district growing at a slower rate than the country as a whole, which has the world's highest fertility rate. The city is located in a pearl millet growing region, while manufacturing industries include bricks, ceramic goods, cement, and weaving. History Niamey was probably founded in the 18th century and originated as a cluster of small villages (Gaweye, Kalley, Maourey, Zongo and Foulani Koira). Niamey was of little importance until the French developed it as a colonial centre in the late 1890s. The town, then with an estimated population of some 1,800, was chosen as the capital of the newly created Military Territory of Niger in 1905, however, the capital was shifted to the more established city of Zinder in 1912. Zinder's proximity to the Nigerian border and distance from French-controlled ports prompted the French to move the capital back to Niamey in 1926, by which time the city had some 3,000 inhabitants. A series of devastating droughts prompted significant population growth during this period, and by 1945 the population was about 8,000. Prior to 1926-27 the Upper Volta-Niger border ran along the Niger river, meaning that Niamey lay directly on the boundary. At the time of independence in 1960 the population had grown to around 30,000. The period from 1970 to 1988 was one in
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Who was the man behind The Chipmunks??
-A: David Seville
-
-Q: Which Lloyd Webber musical premiered in the US on 10th December 1993??
-A: Sunset Boulevard
-
-Q: Who was the next British Prime Minister after Arthur Balfour??
-A: Campbell-Bannerman
-
-Q: Who had a 70s No 1 hit with Kiss You All Over??
-A: Exile
-
-Q: What claimed the life of singer Kathleen Ferrier??
-A: Cancer
-
-Q: Of which African country is Niamey the capital?
-A: 
-==================================================
-
-==================================================
-PROMPT #16:
-==================================================
-James Walter McCord Jr. (January 26, 1924 – June 15, 2017) was an American CIA officer, later head of security for President Richard Nixon's 1972 reelection campaign. He was involved as an electronics expert in the burglaries which precipitated the Watergate scandal. Career McCord was born in Waurika, Oklahoma. He served as a bombardier with the rank of second lieutenant in the Army Air Forces during World War II. He briefly attended Baylor University before receiving a B.B.A. from the University of Texas at Austin in 1949. In 1965, he received an M.S. in international affairs from George Washington University. After beginning his career at the Federal Bureau of Investigation (FBI), McCord worked for the Central Intelligence Agency (CIA), ultimately ascending to the GS-15 directorship of the Agency's Office of Security. For a period of time, he was in charge of physical security at the Agency's Langley headquarters. L. Fletcher Prouty, a former colonel in the United States Air Force, claimed then-Director of Central Intelligence Allen Dulles introduced McCord to him as "my top man.". In 1961, under his direction, a counter-intelligence program was launched against the Fair Play for Cuba Committee. He also held the rank of lieutenant colonel in the United States Air Force Reserve. Watergate scandal Shortly after resigning from the CIA, McCord was interviewed and then hired by Jack Caulfield in January 1972 "for strict, solely defensive security work at the Republican National Committee (RNC) and the Committee to Re-Elect the President (CRP)." Some of the money from this contract came fromadministration as assistant director of the Bureau of the Budget, devoting most of his time to Defense matters. In 1971, President Nixon appointed Schlesinger a member of the Atomic Energy Commission (AEC) and designated him as chairman. Serving in this position for about a year and a half, Schlesinger instituted extensive organizational and management changes in an effort to improve the AEC's regulatory performance. CIA Director Schlesinger was CIA Director from February 2, 1973, to July 2, 1973. He was succeeded by William Colby. Schlesinger was extremely unpopular with CIA staff, as he reduced CIA staff by 7%, and was considered a Nixon loyalist seeking to make the agency more obedient to Nixon. He had a CCTV camera installed near his official portrait at the CIA headquarters in Langley, Va., as it was believed that vandalism of the portrait by disgruntled staff was likely. Secretary of Defense (1973–1975) Schlesinger left the CIA to become Secretary of Defense on July 2, aged 44. As a university professor, researcher at Rand, and government official in three agencies, he had acquired an impressive resume in national security affairs. Nuclear strategy Shortly after assuming office, Schlesinger outlined the basic objectives that would guide his administration: maintain a "strong defense establishment"; "assure the military balance so necessary to deterrence and a more enduring peace"; obtain for members of the military "the respect, dignity and support that are their due"; assume "an . . . obligation to use our citizens' resources wisely"; and "become increasingly competitive with potential adversaries.... [W]e must nota conventional North Vietnamese assault in 1975. The CORDS model and its approach influenced U.S. strategy and thinking on counterinsurgency in the 2000s in Iraq and Afghanistan. CIA HQ: Director Colby returned to Washington in July 1971 and became executive director of CIA. After long-time DCI Richard Helms was dismissed by President Nixon in 1973, James Schlesinger assumed the helm at the Agency. A strong believer in reform of the CIA and the intelligence community more broadly, Schlesinger had written a 1971 Bureau of the Budget report outlining his views on the subject. Colby, who had had a somewhat unorthodox career in the CIA focused on political action and counterinsurgency, agreed with Schlesinger's reformist approach. Schlesinger appointed him head of the clandestine branch in early 1973. When Nixon reshuffled his agency heads and made Schlesinger secretary of defense, Colby emerged as a natural candidate for DCI—apparently on the basis of the recommendation that he was a professional who would not make waves. Colby was known as a media-friendly CIA director. His tenure as DCI, which lasted two and a half tumultuous years, was overshadowed by the Church and Pike congressional investigations into alleged U.S. intelligence malfeasance over the preceding 25 years, including 1975, the so-called Year of Intelligence. Colby's time as DCI was also eventful on the world stage. Shortly after he assumed leadership, the Yom Kippur War broke out, an event that surprised not only the American intelligence agencies but also the Israelis. This intelligence surprise reportedly affected Colby's credibility with the Nixon administration. Colby
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Who was the man behind The Chipmunks??
-A: David Seville
-
-Q: Which Lloyd Webber musical premiered in the US on 10th December 1993??
-A: Sunset Boulevard
-
-Q: Who was the next British Prime Minister after Arthur Balfour??
-A: Campbell-Bannerman
-
-Q: Who had a 70s No 1 hit with Kiss You All Over??
-A: Exile
-
-Q: What claimed the life of singer Kathleen Ferrier??
-A: Cancer
-
-Q: Who was the director of the CIA from 1976-81?
-A: 
-==================================================
-
-==================================================
-PROMPT #17:
-==================================================
-"On the Street Where You Live" is a song with music by Frederick Loewe and lyrics by Alan Jay Lerner from the 1956 Broadway musical My Fair Lady. It is sung in the musical by the character Freddy Eynsford-Hill, who was portrayed by John Michael King in the original production. In the 1964 film version, it was sung by Bill Shirley, dubbing for actor Jeremy Brett. Recorded versions The most popular single of the song was recorded by Vic Damone in 1956 for Columbia Records. It reached No. 4 on the Billboard chart and No. 6 on Cashbox magazine's chart. It was a No. 1 hit in the UK Singles Chart in 1958. Eddie Fisher also had a top 20 Billboard hit with the song in 1956, reaching No. 18. Lawrence Welk and His Orchestra released a version that went to No. 96 in 1956. Andy Williams' recording appeared in the Billboard top 40 in 1964, reaching No. 3 on the adult contemporary chart and No. 28 on the Billboard Hot 100. The song has been recorded by a wide variety of other performers, including Ray Conniff and Bing Crosby, who recorded the song in 1956 for use on his radio show and it was subsequently included in the boxed set The Bing Crosby CBS Radio Recordings (1954–56) issued by Mosaic Records (catalog MD7-245) in 2009, Lawrence Welk (whose band also performed it on his weekly TV series numerous times), Shirley Horn, Doris Day, George Shearing, Frank Chacksfield, Alfie Boe, Bobby Darin, Dean Martin, Mario Lanza,The Times praised it as "Alan Jay Lerner's terrific autobiography". The Street Where I Live was reissued in 1989 by Columbus Books and in 1994 by the Da Capo Press. In 2000, BBC radio broadcast a serialization of the book, read by Henry Goodman, which The Times called "one of the delights of the evening schedule". References Sources Non-fiction books about musical theatre"On the Street Where You Live" is a song from the 1956 Broadway musical My Fair Lady. On the Street Where You Live may also refer to: On the Street Where You Live (TV series), an Irish documentary television series On The Street Where You Live, a 2001 novel by Mary Higgins Clark
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Who was the man behind The Chipmunks??
-A: David Seville
-
-Q: Which Lloyd Webber musical premiered in the US on 10th December 1993??
-A: Sunset Boulevard
-
-Q: Who was the next British Prime Minister after Arthur Balfour??
-A: Campbell-Bannerman
-
-Q: Who had a 70s No 1 hit with Kiss You All Over??
-A: Exile
-
-Q: What claimed the life of singer Kathleen Ferrier??
-A: Cancer
-
-Q: Which musical featured the song The Street Where You Live?
-A: 
-==================================================
-
-==================================================
-PROMPT #18:
-==================================================
-engineers were ordered to end construction work. The Allies were unaware of this and mounted further attacks on the site as part of the United States Army Air Forces experimental Operation Aphrodite, involving radio-controlled B-24 Liberators packed with explosives. Two such attacks were mounted but failed; in the second such attack, on 12 August, Lt Joseph P. Kennedy, Jr. – the elder brother of future US President John F. Kennedy – was killed when the drone aircraft exploded prematurely. By the end of the bombing campaign, over 4,100 tons of bombs had been dropped on Mimoyecques, more than on any other V-weapons site. The Mimoyecques site was never formally abandoned, but German forces left it at the start of September 1944 as the Allies advanced northeast from Normandy towards the Pas de Calais. It was captured on 5 September by the Canadian 3rd Infantry Division. Subsequent investigations and attempted demolition In September 1944, Duncan Sandys ordered the constitution of a Technical Inter-Services Mission under Colonel T.R.B. Sanders. It was given the task of investigating the V-weapons sites at Mimoyecques, Siracourt, Watten, and Wizernes, collectively known to the Allies as the "Heavy Crossbow" sites. Sanders' report was submitted to the War Cabinet on 19 March 1945. Even at this stage the true purpose of the site was unclear. Claims that it had been intended to be used for "electro-magnetic projectors" (railguns), firing huge shells at London, were debunked by Lord Cherwell, Winston Churchill's scientific adviser, who calculated that it would take sixty times the output of Battersearesearched at a facility in Peenemünde along with the V-1 flying bomb. The V-2's first target was Paris on 8 September 1944. The program while advanced proved to be an impediment to the war economy. The large capital investment was not repaid in military effectiveness. The rockets were built at an underground factory at Mittelwerk. Labor to build the A4 rockets came from the Mittelbau-Dora concentration camp. Of the 60,000 people who ended up at the camp 20,000 died, due to the appalling conditions. On 14 April 1944, Speer lost control of Organisation Todt to his Deputy, Franz Xaver Dorsch. He opposed the assassination attempt against Hitler on 20 July 1944. He was not involved in the plot, and played a minor role in the regime's efforts to regain control over Berlin after Hitler survived. After the plot Speer's rivals attacked some of his closest allies and his management system fell out of favor with radicals in the party. He lost yet more authority. Defeat of Nazi Germany Losses of territory and a dramatic expansion of the Allied strategic bombing campaign caused the collapse of the German economy from late 1944. Air attacks on the transport network were particularly effective, as they cut the main centres of production off from essential coal supplies. In January 1945, Speer told Goebbels that armaments production could be sustained for at least a year. However, he concluded that the war was lost after Soviet forces captured the important Silesian industrial region later that month. Nevertheless, Speer believed that Germany shouldof 1944 the Allies continued their gains in the Mediterranean Theatre and massed men and materiel for a European invasion along the French channel coastline. The conspirators began to organize for another attempt to assassinate Hitler and take over both German civil government and its military. The von Stauffenberg bomb attempt and aftermath By the summer of 1944 unrest in the German military and diplomatic ranks was widespread. The Allied landing at Normandy in June and failed German response raised the specter of doom among the upper ranks even of German field marshals. The Schwarze Kapelle responded by organizing a deadly attempt on Hitler's life at his Wolf's Lair compound in East Prussia. Undertaken by an aristocratic member of a hereditarily military family, Colonel Claus von Stauffenberg, the July 20 Plot nearly succeeded. Although surrounded by fatalities from the bomb Hitler escaped with a concussion and various injuries. In the aftermath he was determined to get vengeance upon the plotters. The Gestapo rounded up the members of the Schwarze Kapelle and many, many more it believed were either implicated in or sympathetic to it; according to its records it put 7,000 of them to death. Stauffenberg and three others were summarily shot that night. Most of the conspirators were put on trial in the Volksgerichtshof (People's Court) between August 1944 to February 1945. Many were executed the day after their convictions by hanging from meat hooks at Plötzensee Prison. Architect of the 1943 bomb plot on Hitler's plane Fabian von Schlabrendorff only escaped death because an
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Who was the man behind The Chipmunks??
-A: David Seville
-
-Q: Which Lloyd Webber musical premiered in the US on 10th December 1993??
-A: Sunset Boulevard
-
-Q: Who was the next British Prime Minister after Arthur Balfour??
-A: Campbell-Bannerman
-
-Q: Who had a 70s No 1 hit with Kiss You All Over??
-A: Exile
-
-Q: What claimed the life of singer Kathleen Ferrier??
-A: Cancer
-
-Q: "Who was the target of the failed ""Bomb Plot"" of 1944?"
-A: 
-==================================================
-
-==================================================
-PROMPT #19:
-==================================================
-propelling him into the first rank of international superstars. The album contained the number-one hit "All Night Long", a Caribbean-flavored dance number that was promoted by a colorful music video produced by former Monkee Michael Nesmith. In 1984, he performed "All Night Long" at the ending ceremony of the XXIII Olympic Games in Los Angeles. Several more Top 10 hits followed, the most successful of which was the ballad "Hello" (1984), a sentimental love song that showed how far he had moved from his R&B roots. Richie had three more top ten hits in 1984, "Stuck on You" (No. 3), "Running with the Night" (No. 7) and "Penny Lover" (No. 8), as well as writing and producing "Missing You" for former labelmate and duet partner Diana Ross (No. 10 Pop, No. 1 R&B). In 1985, he wrote and performed "Say You, Say Me" for the film White Nights. The song won an Academy Award and reached No. 1 on the U.S. charts, staying there for four weeks, making it the number-two song of 1986 according to Billboards Year-End Hot 100 chart, behind the charity single "That's What Friends Are For" by Dionne and Friends. He also collaborated with Michael Jackson on the charity single "We Are the World" by USA for Africa, another number-one hit. In 1986, Richie released Dancing on the Ceiling, his last widely popular album, which produced a run of five US and UK hits, "Say You, Say Me" (U.S. No. 1), "Dancing on the Ceiling" (U.S. No. 2), "Love Will Conquer All"top 20 US R&B chart hit in 1972. Their first few recordings were released on Buddah Records, including "Hold Back the Night", which was a hit on the Billboard R&B chart in 1973, before a re-release saw it climb in the UK two years later. Several R&B hits followed during a stay with Philadelphia International subsidiary Golden Fleece (run by Baker-Harris-Young) before they signed to Atlantic Records. Their single "Disco Inferno" (1976), which was included on the Grammy Award-winning Saturday Night Fever: The Original Movie Sound Track in 1977, reached No. 11 on the Billboard Hot 100 chart in May 1978. Other major hits included "Hold Back the Night" (1975) (UK No. 5) and "That's Where the Happy People Go" (1976). In late 1977, the Trammps released the song "The Night the Lights Went Out" to commemorate the electrical blackout that affected New York City on July 13–14, 1977. Their signature song "Disco Inferno" has been covered by Tina Turner and Cyndi Lauper. In addition, Graham Parker covered "Hold Back the Night" on "The Pink Parker EP" in 1977, and reached No. 24 in the UK Singles Chart, and top 60 in the US. In 2021, "Disco Inferno" was certified Silver by the British Phonographic Industry, together with "Can We Come Together" (from the album Where the Happy People Go). Dissolution and aftermath On September 19, 2005, the group's "Disco Inferno" was inducted into the Dance Music Hall of Fame at a ceremony held in New York. The song was part-written by Ron Kersey, a producer-arranger"Hold On to the Nights" is a power ballad written and performed by American rock singer/songwriter/musician Richard Marx. This was the fourth and final single released from his self-titled debut album, and his first to reach number one on the US Billboard Hot 100 chart. The song has been re-released on numerous albums and is included on Marx's live performance DVD A Night Out with Friends (2012). Release "Hold On to the Nights" reached the Billboard Hot 100 number 1 position on July 23, 1988, preventing Def Leppard's "Pour Some Sugar on Me" from reaching the top spot that same week. The song was on the chart for twenty-one weeks, and left the chart at number 91. The song also reached at number three on the Billboard Adult Contemporary chart. Chart performance Charts Personnel Richard Marx – vocals, keyboards, acoustic piano Michael Landau – guitars Patrick O'Hearn – bass Tris Imboden – drums Paulinho da Costa – percussion Other performances Marx appeared as lounge singer/piano player Buddy Daquiri in the "Poison Fire Teats Universe" episode of the TV series Life in Pieces in 2017, in which he played the song on the piano while whistling. References 1987 songs 1988 singles Richard Marx songs Billboard Hot 100 number-one singles Songs written by Richard Marx Pop ballads Rock ballads EMI Records singles Songs about nights
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Who was the man behind The Chipmunks??
-A: David Seville
-
-Q: Which Lloyd Webber musical premiered in the US on 10th December 1993??
-A: Sunset Boulevard
-
-Q: Who was the next British Prime Minister after Arthur Balfour??
-A: Campbell-Bannerman
-
-Q: Who had a 70s No 1 hit with Kiss You All Over??
-A: Exile
-
-Q: What claimed the life of singer Kathleen Ferrier??
-A: Cancer
-
-Q: Who had an 80s No 1 hit with Hold On To The Nights?
-A: 
-==================================================
-
-==================================================
-PROMPT #20:
-==================================================
-Turner Classic Movies in November 2006 features directors Steven Spielberg, Clint Eastwood, and Martin Scorsese, who suggest that the string of classic films Ford directed during 1936 to 1941 was due in part to an intense six-month extramarital affair with Katharine Hepburn, the star of Mary of Scotland (1936), an Elizabethan costume drama. 1939–1941 Stagecoach (1939) was Ford's first western since 3 Bad Men in 1926, and it was his first with sound. Orson Welles claimed that he watched Stagecoach forty times in preparation for making Citizen Kane. It remains one of the most admired and imitated of all Hollywood movies, not least for its climactic stagecoach chase and the hair-raising horse-jumping scene, performed by the stuntman Yakima Canutt. The Dudley Nichols–Ben Hecht screenplay was based on an Ernest Haycox story that Ford had spotted in Collier's magazine and he purchased the screen rights for just $2500. Production chief Walter Wanger urged Ford to hire Gary Cooper and Marlene Dietrich for the lead roles, but eventually accepted Ford's decision to cast Claire Trevor as Dallas and a virtual unknown, his friend John Wayne, as Ringo; Wanger reportedly had little further influence over the production. In making Stagecoach, Ford faced entrenched industry prejudice about the now-hackneyed genre which he had helped to make so popular. Although low-budget western features and serials were still being churned out in large numbers by "Poverty Row" studios, the genre had fallen out of favor with the big studios during the 1930s and they were regarded as B-grade "pulp" movies at best.Stagecoach is a 1986 American made-for-television Western action drama film and remake of the classic 1939 film Stagecoach, directed by Ted Post and starring Kris Kristofferson as the Ringo Kid, the role originally played by John Wayne. Willie Nelson portrays famous gunslinger and dentist Doc Holliday, Johnny Cash portrays Marshal Curly Wilcox and Waylon Jennings plays the gambler Hatfield. The four main stars of the film (Nelson, Kristofferson, Cash and Jennings) were associated as members of the country music supergroup The Highwaymen. The supporting cast features Elizabeth Ashley, Anthony Newley, Tony Franciosa, Mary Crosby, June Carter Cash and Jessi Colter. Plot In 1880, a group of strangers boards the east-bound stagecoach from Tonto, Arizona Territory, to Lordsburg, New Mexico Territory. The travelers seem ordinary, but many have secrets from which they are running. Among them are Dallas, a prostitute, who is being driven out of town; an alcoholic dentist, Doc Holliday; pregnant Lucy Mallory, who is meeting her cavalry officer husband; and whiskey salesman Trevor Peacock. As the stage sets out, U.S. Cavalry Lieutenant Blanchard announces that Geronimo and his Apaches are on the warpath; his small troop will provide an escort to Dry Fork. Cast Willie Nelson as Doc Holliday Kris Kristofferson as Ringo / Ringo Kid / Bill Williams Johnny Cash as Marshal Curly Wilcox Waylon Jennings as Hatfield (Gambler) John Schneider as Buck (Overland Stage Driver) Elizabeth Ashley as Dallas Anthony Newley as Trevor Peacock (Old John's Whiskey Salesman) Tony Franciosa as Henry Gatewood (Tonto Banker) Merritt Butrick as Lieutenant Blanchard Mary CrosbyStagecoach is a 1939 American Western film directed by John Ford and starring Claire Trevor and John Wayne in his breakthrough role. The screenplay by Dudley Nichols is an adaptation of "The Stage to Lordsburg", a 1937 short story by Ernest Haycox. The film follows a group of strangers riding on a stagecoach through dangerous Apache territory. The film has long been recognized as an important work that transcends the Western genre. Philosopher Robert B. Pippin has observed that both the collection of characters and their journey "are archetypal rather than merely individual" and that the film is a "mythic representation of the American aspiration toward a form of politically meaningful equality." In 1995, the film was deemed "culturally, historically, or aesthetically significant" by the United States Library of Congress and selected for preservation in their National Film Registry. Still, Stagecoach has not avoided controversy. Like most Westerns of the era, its depiction of Native Americans as simplistic savages has been criticized. Stagecoach was the first of many Westerns that Ford shot in Monument Valley, on the Arizona–Utah border in the American Southwest. Many of the movies Ford shot there also starred John Wayne. Scenes from Stagecoach, including a sequence introducing John Wayne's character the Ringo Kid, blended shots of Monument Valley with shots filmed on the Iverson Movie Ranch in Chatsworth, California, RKO Encino Movie Ranch, and other locations. Geographic incongruities are visible throughout the film, including the closing scene where Ringo (Wayne) and Dallas (Trevor) depart Lordsburg, in southwestern New Mexico, by way of
-Think hard, but answer shortly and concisely. Only give direct answers to the questions. No additional explanations. Directly answer these questions:
-Q: Who was the man behind The Chipmunks??
-A: David Seville
-
-Q: Which Lloyd Webber musical premiered in the US on 10th December 1993??
-A: Sunset Boulevard
-
-Q: Who was the next British Prime Minister after Arthur Balfour??
-A: Campbell-Bannerman
-
-Q: Who had a 70s No 1 hit with Kiss You All Over??
-A: Exile
-
-Q: What claimed the life of singer Kathleen Ferrier??
-A: Cancer
-
-Q: Who directed the classic 30s western Stagecoach?
-A: 
-==================================================
-

benchmarks/enron_emails/README.md

@@ -0,0 +1,141 @@
+# Enron Emails Benchmark
+
+A comprehensive RAG benchmark for evaluating LEANN search and generation on the Enron email corpus. It mirrors the structure and CLI of the existing FinanceBench and LAION benches, using stage-based evaluation with Recall@3 and generation timing.
+
+- Dataset: Enron email CSV (e.g., Kaggle wcukierski/enron-email-dataset) for passages
+- Queries: corbt/enron_emails_sample_questions (filtered for realistic questions)
+- Metrics: Recall@3 vs FAISS Flat baseline + Generation evaluation with Qwen3-8B
+
+## Layout
+
+benchmarks/enron_emails/
+- setup_enron_emails.py: Prepare passages, build LEANN index, build FAISS baseline
+- evaluate_enron_emails.py: Evaluate retrieval recall (Stages 2-5) + generation with Qwen3-8B
+- data/: Generated passages, queries, embeddings-related files
+- baseline/: FAISS Flat baseline files
+- llm_utils.py: LLM utilities for Qwen3-8B generation (in parent directory)
+
+## Quickstart
+
+1) Prepare the data and index
+
+cd benchmarks/enron_emails
+python setup_enron_emails.py --data-dir data
+
+Notes:
+- If `--emails-csv` is omitted, the script attempts to download from Kaggle dataset `wcukierski/enron-email-dataset` using Kaggle API (requires `KAGGLE_USERNAME` and `KAGGLE_KEY`).
+  Alternatively, pass a local path to `--emails-csv`.
+
+Notes:
+- The script parses emails, chunks header/body into passages, builds a compact LEANN index, and then builds a FAISS Flat baseline from the same passages and embedding model.
+- Optionally, it will also create evaluation queries from HuggingFace dataset `corbt/enron_emails_sample_questions`.
+
+2) Run recall evaluation (Stage 2)
+
+python evaluate_enron_emails.py --index data/enron_index_hnsw.leann --stage 2
+
+3) Complexity sweep (Stage 3)
+
+python evaluate_enron_emails.py --index data/enron_index_hnsw.leann --stage 3 --target-recall 0.90 --max-queries 200
+
+Stage 3 uses binary search over complexity to find the minimal value achieving the target Recall@3 (assumes recall is non-decreasing with complexity). The search expands the upper bound as needed and snaps complexity to multiples of 8.
+
+4) Index comparison (Stage 4)
+
+python evaluate_enron_emails.py --index data/enron_index_hnsw.leann --stage 4 --complexity 88 --max-queries 100 --output results.json
+
+5) Generation evaluation (Stage 5)
+
+python evaluate_enron_emails.py --index data/enron_index_hnsw.leann --stage 5 --complexity 88 --llm-backend hf --model-name Qwen/Qwen3-8B
+
+6) Combined index + generation evaluation (Stages 4+5, recommended)
+
+python evaluate_enron_emails.py --index data/enron_index_hnsw.leann --stage 45 --complexity 88 --llm-backend hf
+
+Notes:
+- Minimal CLI: you can run from repo root with only `--index`, defaults match financebench/laion patterns:
+  - `--stage` defaults to `all` (runs 2, 3, 4, 5)
+  - `--baseline-dir` defaults to `baseline`
+  - `--queries` defaults to `data/evaluation_queries.jsonl` (or falls back to the index directory)
+  - `--llm-backend` defaults to `hf` (HuggingFace), can use `vllm`
+  - `--model-name` defaults to `Qwen/Qwen3-8B`
+- Fail-fast behavior: no silent fallbacks. If compact index cannot run with recompute, it errors out.
+- Stage 5 requires Stage 4 retrieval results. Use `--stage 45` to run both efficiently.
+
+Optional flags:
+- --queries data/evaluation_queries.jsonl (custom queries file)
+- --baseline-dir baseline (where FAISS baseline lives)
+- --complexity 88 (LEANN complexity parameter, optimal for 90% recall)
+- --llm-backend hf|vllm (LLM backend for generation)
+- --model-name Qwen/Qwen3-8B (LLM model for generation)
+- --max-queries 1000 (limit number of queries for evaluation)
+
+## Files Produced
+- data/enron_passages_preview.jsonl: Small preview of passages used (for inspection)
+- data/enron_index_hnsw.leann.*: LEANN index files
+- baseline/faiss_flat.index + baseline/metadata.pkl: FAISS baseline with passage IDs
+- data/evaluation_queries.jsonl: Query file (id + query; includes GT IDs for reference)
+
+## Notes
+- Evaluates both retrieval Recall@3 and generation timing with Qwen3-8B thinking model.
+- The emails CSV must contain a column named "message" (raw RFC822 email) and a column named "file" for source identifier. Message-ID headers are parsed as canonical message IDs when present.
+- Qwen3-8B requires special handling for thinking models with chat templates and <think></think> tag processing.
+
+## Stages Summary
+
+- Stage 2 (Recall@3):
+  - Compares LEANN vs FAISS Flat baseline on Recall@3.
+  - Compact index runs with `recompute_embeddings=True`.
+
+- Stage 3 (Binary Search for Complexity):
+  - Builds a non-compact index (`<index>_noncompact.leann`) and runs binary search with `recompute_embeddings=False` to find the minimal complexity achieving target Recall@3 (default 90%).
+
+- Stage 4 (Index Comparison):
+  - Reports .index-only sizes for compact vs non-compact.
+  - Measures timings on queries by default: non-compact (no recompute) vs compact (with recompute).
+  - Stores retrieval results for Stage 5 generation evaluation.
+  - Fails fast if compact recompute cannot run.
+  - If `--complexity` is not provided, the script tries to use the best complexity from Stage 3:
+    - First from the current run (when running `--stage all`), otherwise
+    - From `enron_stage3_results.json` saved next to the index during the last Stage 3 run.
+    - If neither exists, Stage 4 will error and ask you to run Stage 3 or pass `--complexity`.
+
+- Stage 5 (Generation Evaluation):
+  - Uses Qwen3-8B thinking model for RAG generation on retrieved documents from Stage 4.
+  - Supports HuggingFace (`hf`) and vLLM (`vllm`) backends.
+  - Measures generation timing separately from search timing.
+  - Requires Stage 4 results (no additional searching performed).
+
+## Example Results
+
+These are sample results obtained on Enron data using all-mpnet-base-v2 and Qwen3-8B.
+
+- Stage 3 (Binary Search):
+  - Minimal complexity achieving 90% Recall@3: 88
+  - Sampled points:
+    - C=8 → 59.9% Recall@3
+    - C=72 → 89.4% Recall@3
+    - C=88 → 90.2% Recall@3
+    - C=96 → 90.7% Recall@3
+    - C=112 → 91.1% Recall@3
+    - C=136 → 91.3% Recall@3
+    - C=256 → 92.0% Recall@3
+
+- Stage 4 (Index Sizes, .index only):
+  - Compact: ~2.2 MB
+  - Non-compact: ~82.0 MB
+  - Storage saving by compact: ~97.3%
+
+- Stage 4 (Search Timing, 988 queries, complexity=88):
+  - Non-compact (no recompute): ~0.0075 s avg per query
+  - Compact (with recompute): ~1.981 s avg per query
+  - Speed ratio (non-compact/compact): ~0.0038x
+
+- Stage 5 (RAG Generation, 988 queries, Qwen3-8B):
+  - Average generation time: ~22.302 s per query
+  - Total queries processed: 988
+  - LLM backend: HuggingFace transformers
+  - Model: Qwen/Qwen3-8B (thinking model with <think></think> processing)
+
+Full JSON output is saved by the script (see `--output`), e.g.:
+`benchmarks/enron_emails/results_enron_stage45.json`.

benchmarks/enron_emails/data/.gitignore

@@ -0,0 +1 @@
+downloads/

benchmarks/enron_emails/evaluate_enron_emails.py

@@ -0,0 +1,614 @@
+"""
+Enron Emails Benchmark Evaluation - Retrieval Recall@3 (Stages 2/3/4)
+Follows the style of FinanceBench/LAION: Stage 2 recall vs FAISS baseline,
+Stage 3 complexity sweep to target recall, Stage 4 index comparison.
+On errors, fail fast without fallbacks.
+"""
+
+import argparse
+import json
+import logging
+import os
+import pickle
+from pathlib import Path
+
+import numpy as np
+from leann import LeannBuilder, LeannSearcher
+from leann_backend_hnsw import faiss
+
+from ..llm_utils import generate_hf, generate_vllm, load_hf_model, load_vllm_model
+
+# Setup logging to reduce verbose output
+logging.basicConfig(level=logging.WARNING)
+logging.getLogger("leann.api").setLevel(logging.WARNING)
+logging.getLogger("leann_backend_hnsw").setLevel(logging.WARNING)
+
+
+class RecallEvaluator:
+    """Stage 2: Evaluate Recall@3 (LEANN vs FAISS)"""
+
+    def __init__(self, index_path: str, baseline_dir: str):
+        self.index_path = index_path
+        self.baseline_dir = baseline_dir
+        self.searcher = LeannSearcher(index_path)
+
+        baseline_index_path = os.path.join(baseline_dir, "faiss_flat.index")
+        metadata_path = os.path.join(baseline_dir, "metadata.pkl")
+
+        self.faiss_index = faiss.read_index(baseline_index_path)
+        with open(metadata_path, "rb") as f:
+            self.passage_ids = pickle.load(f)
+
+        print(f"📚 Loaded FAISS flat baseline with {self.faiss_index.ntotal} vectors")
+
+        # No fallbacks here; if embedding server is needed but fails, the caller will see the error.
+
+    def evaluate_recall_at_3(
+        self, queries: list[str], complexity: int = 64, recompute_embeddings: bool = True
+    ) -> float:
+        """Evaluate recall@3 using FAISS Flat as ground truth"""
+        from leann.api import compute_embeddings
+
+        recompute_str = "with recompute" if recompute_embeddings else "no recompute"
+        print(f"🔍 Evaluating recall@3 with complexity={complexity} ({recompute_str})...")
+
+        total_recall = 0.0
+        for i, query in enumerate(queries):
+            # Compute query embedding with the same model/mode as the index
+            q_emb = compute_embeddings(
+                [query],
+                self.searcher.embedding_model,
+                mode=self.searcher.embedding_mode,
+                use_server=False,
+            ).astype(np.float32)
+
+            # Search FAISS Flat ground truth
+            n = q_emb.shape[0]
+            k = 3
+            distances = np.zeros((n, k), dtype=np.float32)
+            labels = np.zeros((n, k), dtype=np.int64)
+            self.faiss_index.search(
+                n,
+                faiss.swig_ptr(q_emb),
+                k,
+                faiss.swig_ptr(distances),
+                faiss.swig_ptr(labels),
+            )
+
+            baseline_ids = {self.passage_ids[idx] for idx in labels[0]}
+
+            # Search with LEANN (may require embedding server depending on index configuration)
+            results = self.searcher.search(
+                query,
+                top_k=3,
+                complexity=complexity,
+                recompute_embeddings=recompute_embeddings,
+            )
+            test_ids = {r.id for r in results}
+
+            intersection = test_ids.intersection(baseline_ids)
+            recall = len(intersection) / 3.0
+            total_recall += recall
+
+            if i < 3:
+                print(f"  Q{i + 1}: '{query[:60]}...' -> Recall@3: {recall:.3f}")
+                print(f"    FAISS: {list(baseline_ids)}")
+                print(f"    LEANN: {list(test_ids)}")
+                print(f"    ∩: {list(intersection)}")
+
+        avg = total_recall / max(1, len(queries))
+        print(f"📊 Average Recall@3: {avg:.3f} ({avg * 100:.1f}%)")
+        return avg
+
+    def cleanup(self):
+        if hasattr(self, "searcher"):
+            self.searcher.cleanup()
+
+
+class EnronEvaluator:
+    def __init__(self, index_path: str):
+        self.index_path = index_path
+        self.searcher = LeannSearcher(index_path)
+
+    def load_queries(self, queries_file: str) -> list[str]:
+        queries: list[str] = []
+        with open(queries_file, encoding="utf-8") as f:
+            for line in f:
+                if not line.strip():
+                    continue
+                data = json.loads(line)
+                if "query" in data:
+                    queries.append(data["query"])
+        print(f"📊 Loaded {len(queries)} queries from {queries_file}")
+        return queries
+
+    def cleanup(self):
+        if self.searcher:
+            self.searcher.cleanup()
+
+    def analyze_index_sizes(self) -> dict:
+        """Analyze index sizes (.index only), similar to LAION bench."""
+
+        print("📏 Analyzing index sizes (.index only)...")
+        index_path = Path(self.index_path)
+        index_dir = index_path.parent
+        index_name = index_path.stem
+
+        sizes: dict[str, float] = {}
+        index_file = index_dir / f"{index_name}.index"
+        meta_file = index_dir / f"{index_path.name}.meta.json"
+        passages_file = index_dir / f"{index_path.name}.passages.jsonl"
+        passages_idx_file = index_dir / f"{index_path.name}.passages.idx"
+
+        sizes["index_only_mb"] = (
+            index_file.stat().st_size / (1024 * 1024) if index_file.exists() else 0.0
+        )
+        sizes["metadata_mb"] = (
+            meta_file.stat().st_size / (1024 * 1024) if meta_file.exists() else 0.0
+        )
+        sizes["passages_text_mb"] = (
+            passages_file.stat().st_size / (1024 * 1024) if passages_file.exists() else 0.0
+        )
+        sizes["passages_index_mb"] = (
+            passages_idx_file.stat().st_size / (1024 * 1024) if passages_idx_file.exists() else 0.0
+        )
+
+        print(f"  📁 .index size: {sizes['index_only_mb']:.1f} MB")
+        return sizes
+
+    def create_non_compact_index_for_comparison(self, non_compact_index_path: str) -> dict:
+        """Create a non-compact index for comparison using current passages and embeddings."""
+
+        current_index_path = Path(self.index_path)
+        current_index_dir = current_index_path.parent
+        current_index_name = current_index_path.name
+
+        # Read metadata to get passage source and embedding model
+        meta_path = current_index_dir / f"{current_index_name}.meta.json"
+        with open(meta_path, encoding="utf-8") as f:
+            meta = json.load(f)
+
+        passage_source = meta["passage_sources"][0]
+        passage_file = passage_source["path"]
+
+        # Convert relative path to absolute
+        if not Path(passage_file).is_absolute():
+            passage_file = current_index_dir / Path(passage_file).name
+
+        # Load all passages and ids
+        ids: list[str] = []
+        texts: list[str] = []
+        with open(passage_file, encoding="utf-8") as f:
+            for line in f:
+                if line.strip():
+                    data = json.loads(line)
+                    ids.append(str(data["id"]))
+                    texts.append(data["text"])
+
+        # Compute embeddings using the same method as LEANN
+        from leann.api import compute_embeddings
+
+        embeddings = compute_embeddings(
+            texts,
+            meta["embedding_model"],
+            mode=meta.get("embedding_mode", "sentence-transformers"),
+            use_server=False,
+        ).astype(np.float32)
+
+        # Build non-compact index with same passages and embeddings
+        builder = LeannBuilder(
+            backend_name="hnsw",
+            embedding_model=meta["embedding_model"],
+            embedding_mode=meta.get("embedding_mode", "sentence-transformers"),
+            is_recompute=False,
+            is_compact=False,
+            **{
+                k: v
+                for k, v in meta.get("backend_kwargs", {}).items()
+                if k not in ["is_recompute", "is_compact"]
+            },
+        )
+
+        # Persist a pickle for build_index_from_embeddings
+        pkl_path = current_index_dir / f"{Path(non_compact_index_path).stem}_embeddings.pkl"
+        with open(pkl_path, "wb") as pf:
+            pickle.dump((ids, embeddings), pf)
+
+        print(
+            f"🔨 Building non-compact index at {non_compact_index_path} from precomputed embeddings..."
+        )
+        builder.build_index_from_embeddings(non_compact_index_path, str(pkl_path))
+
+        # Analyze the non-compact index size
+        temp_evaluator = EnronEvaluator(non_compact_index_path)
+        non_compact_sizes = temp_evaluator.analyze_index_sizes()
+        non_compact_sizes["index_type"] = "non_compact"
+
+        return non_compact_sizes
+
+    def compare_index_performance(
+        self, non_compact_path: str, compact_path: str, test_queries: list[str], complexity: int
+    ) -> dict:
+        """Compare search speed for non-compact vs compact indexes."""
+        import time
+
+        results: dict = {
+            "non_compact": {"search_times": []},
+            "compact": {"search_times": []},
+            "avg_search_times": {},
+            "speed_ratio": 0.0,
+            "retrieval_results": [],  # Store retrieval results for Stage 5
+        }
+
+        print("⚡ Comparing search performance between indexes...")
+        # Non-compact (no recompute)
+        print("  🔍 Testing non-compact index (no recompute)...")
+        non_compact_searcher = LeannSearcher(non_compact_path)
+        for q in test_queries:
+            t0 = time.time()
+            _ = non_compact_searcher.search(
+                q, top_k=3, complexity=complexity, recompute_embeddings=False
+            )
+            results["non_compact"]["search_times"].append(time.time() - t0)
+
+        # Compact (with recompute). Fail fast if it cannot run.
+        print("  🔍 Testing compact index (with recompute)...")
+        compact_searcher = LeannSearcher(compact_path)
+        for q in test_queries:
+            t0 = time.time()
+            docs = compact_searcher.search(
+                q, top_k=3, complexity=complexity, recompute_embeddings=True
+            )
+            results["compact"]["search_times"].append(time.time() - t0)
+
+            # Store retrieval results for Stage 5
+            results["retrieval_results"].append(
+                {"query": q, "retrieved_docs": [{"id": doc.id, "text": doc.text} for doc in docs]}
+            )
+        compact_searcher.cleanup()
+
+        if results["non_compact"]["search_times"]:
+            results["avg_search_times"]["non_compact"] = sum(
+                results["non_compact"]["search_times"]
+            ) / len(results["non_compact"]["search_times"])
+        if results["compact"]["search_times"]:
+            results["avg_search_times"]["compact"] = sum(results["compact"]["search_times"]) / len(
+                results["compact"]["search_times"]
+            )
+        if results["avg_search_times"].get("compact", 0) > 0:
+            results["speed_ratio"] = (
+                results["avg_search_times"]["non_compact"] / results["avg_search_times"]["compact"]
+            )
+        else:
+            results["speed_ratio"] = 0.0
+
+        non_compact_searcher.cleanup()
+        return results
+
+    def evaluate_complexity(
+        self,
+        recall_eval: "RecallEvaluator",
+        queries: list[str],
+        target: float = 0.90,
+        c_min: int = 8,
+        c_max: int = 256,
+        max_iters: int = 10,
+        recompute: bool = False,
+    ) -> dict:
+        """Binary search minimal complexity achieving target recall (monotonic assumption)."""
+
+        def round_c(x: int) -> int:
+            # snap to multiple of 8 like other benches typically do
+            return max(1, int((x + 7) // 8) * 8)
+
+        metrics: list[dict] = []
+
+        lo = round_c(c_min)
+        hi = round_c(c_max)
+
+        print(
+            f"🧪 Binary search complexity in [{lo}, {hi}] for target Recall@3>={int(target * 100)}%..."
+        )
+
+        # Ensure upper bound can reach target; expand if needed (up to a cap)
+        r_lo = recall_eval.evaluate_recall_at_3(
+            queries, complexity=lo, recompute_embeddings=recompute
+        )
+        metrics.append({"complexity": lo, "recall_at_3": r_lo})
+        r_hi = recall_eval.evaluate_recall_at_3(
+            queries, complexity=hi, recompute_embeddings=recompute
+        )
+        metrics.append({"complexity": hi, "recall_at_3": r_hi})
+
+        cap = 1024
+        while r_hi < target and hi < cap:
+            lo = hi
+            r_lo = r_hi
+            hi = round_c(hi * 2)
+            r_hi = recall_eval.evaluate_recall_at_3(
+                queries, complexity=hi, recompute_embeddings=recompute
+            )
+            metrics.append({"complexity": hi, "recall_at_3": r_hi})
+
+        if r_hi < target:
+            print(f"⚠️ Max complexity {hi} did not reach target recall {target:.2f}.")
+            print("📈 Observations:")
+            for m in metrics:
+                print(f"  C={m['complexity']:>4} -> Recall@3={m['recall_at_3'] * 100:.1f}%")
+            return {"metrics": metrics, "best_complexity": None, "target_recall": target}
+
+        # Binary search within [lo, hi]
+        best = hi
+        iters = 0
+        while lo < hi and iters < max_iters:
+            mid = round_c((lo + hi) // 2)
+            r_mid = recall_eval.evaluate_recall_at_3(
+                queries, complexity=mid, recompute_embeddings=recompute
+            )
+            metrics.append({"complexity": mid, "recall_at_3": r_mid})
+            if r_mid >= target:
+                best = mid
+                hi = mid
+            else:
+                lo = mid + 8  # move past mid, respecting multiple-of-8 step
+            iters += 1
+
+        print("📈 Binary search results (sampled points):")
+        # Print unique complexity entries ordered by complexity
+        for m in sorted(
+            {m["complexity"]: m for m in metrics}.values(), key=lambda x: x["complexity"]
+        ):
+            print(f"  C={m['complexity']:>4} -> Recall@3={m['recall_at_3'] * 100:.1f}%")
+        print(f"✅ Minimal complexity achieving {int(target * 100)}% recall: {best}")
+        return {"metrics": metrics, "best_complexity": best, "target_recall": target}
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Enron Emails Benchmark Evaluation")
+    parser.add_argument("--index", required=True, help="Path to LEANN index")
+    parser.add_argument(
+        "--queries", default="data/evaluation_queries.jsonl", help="Path to evaluation queries"
+    )
+    parser.add_argument(
+        "--stage",
+        choices=["2", "3", "4", "5", "all", "45"],
+        default="all",
+        help="Which stage to run (2=recall, 3=complexity, 4=index comparison, 5=generation)",
+    )
+    parser.add_argument("--complexity", type=int, default=None, help="LEANN search complexity")
+    parser.add_argument("--baseline-dir", default="baseline", help="Baseline output directory")
+    parser.add_argument(
+        "--max-queries", type=int, help="Limit number of queries to evaluate", default=1000
+    )
+    parser.add_argument(
+        "--target-recall", type=float, default=0.90, help="Target Recall@3 for Stage 3"
+    )
+    parser.add_argument("--output", help="Save results to JSON file")
+    parser.add_argument("--llm-backend", choices=["hf", "vllm"], default="hf", help="LLM backend")
+    parser.add_argument("--model-name", default="Qwen/Qwen3-8B", help="Model name")
+
+    args = parser.parse_args()
+
+    # Resolve queries file: if default path not found, fall back to index's directory
+    if not os.path.exists(args.queries):
+        from pathlib import Path
+
+        idx_dir = Path(args.index).parent
+        fallback_q = idx_dir / "evaluation_queries.jsonl"
+        if fallback_q.exists():
+            args.queries = str(fallback_q)
+
+    baseline_index_path = os.path.join(args.baseline_dir, "faiss_flat.index")
+    if not os.path.exists(baseline_index_path):
+        print(f"❌ FAISS baseline not found at {baseline_index_path}")
+        print("💡 Please run setup_enron_emails.py first to build the baseline")
+        raise SystemExit(1)
+
+    results_out: dict = {}
+
+    if args.stage in ("2", "all"):
+        print("🚀 Starting Stage 2: Recall@3 evaluation")
+        evaluator = RecallEvaluator(args.index, args.baseline_dir)
+
+        enron_eval = EnronEvaluator(args.index)
+        queries = enron_eval.load_queries(args.queries)
+        queries = queries[:10]
+        print(f"🧪 Using first {len(queries)} queries")
+
+        complexity = args.complexity or 64
+        r = evaluator.evaluate_recall_at_3(queries, complexity)
+        results_out["stage2"] = {"complexity": complexity, "recall_at_3": r}
+        evaluator.cleanup()
+        enron_eval.cleanup()
+        print("✅ Stage 2 completed!\n")
+
+    if args.stage in ("3", "all"):
+        print("🚀 Starting Stage 3: Binary search for target recall (no recompute)")
+        enron_eval = EnronEvaluator(args.index)
+        queries = enron_eval.load_queries(args.queries)
+        queries = queries[: args.max_queries]
+        print(f"🧪 Using first {len(queries)} queries")
+
+        # Build non-compact index for fast binary search (recompute_embeddings=False)
+        from pathlib import Path
+
+        index_path = Path(args.index)
+        non_compact_index_path = str(index_path.parent / f"{index_path.stem}_noncompact.leann")
+        enron_eval.create_non_compact_index_for_comparison(non_compact_index_path)
+
+        # Use non-compact evaluator for binary search with recompute=False
+        evaluator_nc = RecallEvaluator(non_compact_index_path, args.baseline_dir)
+        sweep = enron_eval.evaluate_complexity(
+            evaluator_nc, queries, target=args.target_recall, recompute=False
+        )
+        results_out["stage3"] = sweep
+        # Persist default stage 3 results near the index for Stage 4 auto-pickup
+        from pathlib import Path
+
+        default_stage3_path = Path(args.index).parent / "enron_stage3_results.json"
+        with open(default_stage3_path, "w", encoding="utf-8") as f:
+            json.dump({"stage3": sweep}, f, indent=2)
+        print(f"📝 Saved Stage 3 summary to {default_stage3_path}")
+        evaluator_nc.cleanup()
+        enron_eval.cleanup()
+        print("✅ Stage 3 completed!\n")
+
+    if args.stage in ("4", "all", "45"):
+        print("🚀 Starting Stage 4: Index size + performance comparison")
+        evaluator = RecallEvaluator(args.index, args.baseline_dir)
+        enron_eval = EnronEvaluator(args.index)
+        queries = enron_eval.load_queries(args.queries)
+        test_q = queries[: min(args.max_queries, len(queries))]
+
+        current_sizes = enron_eval.analyze_index_sizes()
+        # Build non-compact index for comparison (no fallback)
+        from pathlib import Path
+
+        index_path = Path(args.index)
+        non_compact_path = str(index_path.parent / f"{index_path.stem}_noncompact.leann")
+        non_compact_sizes = enron_eval.create_non_compact_index_for_comparison(non_compact_path)
+        nc_eval = EnronEvaluator(non_compact_path)
+
+        if (
+            current_sizes.get("index_only_mb", 0) > 0
+            and non_compact_sizes.get("index_only_mb", 0) > 0
+        ):
+            storage_saving_percent = max(
+                0.0,
+                100.0 * (1.0 - current_sizes["index_only_mb"] / non_compact_sizes["index_only_mb"]),
+            )
+        else:
+            storage_saving_percent = 0.0
+
+        if args.complexity is None:
+            # Prefer in-session Stage 3 result
+            if "stage3" in results_out and results_out["stage3"].get("best_complexity") is not None:
+                complexity = results_out["stage3"]["best_complexity"]
+                print(f"📥 Using best complexity from Stage 3 in-session: {complexity}")
+            else:
+                # Try to load last saved Stage 3 result near index
+                default_stage3_path = Path(args.index).parent / "enron_stage3_results.json"
+                if default_stage3_path.exists():
+                    with open(default_stage3_path, encoding="utf-8") as f:
+                        prev = json.load(f)
+                    complexity = prev.get("stage3", {}).get("best_complexity")
+                    if complexity is None:
+                        raise SystemExit(
+                            "❌ Stage 4: No --complexity and no best_complexity found in saved Stage 3 results"
+                        )
+                    print(f"📥 Using best complexity from saved Stage 3: {complexity}")
+                else:
+                    raise SystemExit(
+                        "❌ Stage 4 requires --complexity if Stage 3 hasn't been run. Run stage 3 first or pass --complexity."
+                    )
+        else:
+            complexity = args.complexity
+
+        comp = enron_eval.compare_index_performance(
+            non_compact_path, args.index, test_q, complexity=complexity
+        )
+        results_out["stage4"] = {
+            "current_index": current_sizes,
+            "non_compact_index": non_compact_sizes,
+            "storage_saving_percent": storage_saving_percent,
+            "performance_comparison": comp,
+        }
+        nc_eval.cleanup()
+        evaluator.cleanup()
+        enron_eval.cleanup()
+        print("✅ Stage 4 completed!\n")
+
+    if args.stage in ("5", "all"):
+        print("🚀 Starting Stage 5: Generation evaluation with Qwen3-8B")
+
+        # Check if Stage 4 results exist
+        if "stage4" not in results_out or "performance_comparison" not in results_out["stage4"]:
+            print("❌ Stage 5 requires Stage 4 retrieval results")
+            print("💡 Run Stage 4 first or use --stage all")
+            raise SystemExit(1)
+
+        retrieval_results = results_out["stage4"]["performance_comparison"]["retrieval_results"]
+        if not retrieval_results:
+            print("❌ No retrieval results found from Stage 4")
+            raise SystemExit(1)
+
+        print(f"📁 Using {len(retrieval_results)} retrieval results from Stage 4")
+
+        # Load LLM
+        try:
+            if args.llm_backend == "hf":
+                tokenizer, model = load_hf_model(args.model_name)
+
+                def llm_func(prompt):
+                    return generate_hf(tokenizer, model, prompt)
+            else:  # vllm
+                llm, sampling_params = load_vllm_model(args.model_name)
+
+                def llm_func(prompt):
+                    return generate_vllm(llm, sampling_params, prompt)
+
+            # Run generation using stored retrieval results
+            import time
+
+            from llm_utils import create_prompt
+
+            generation_times = []
+            responses = []
+
+            print("🤖 Running generation on pre-retrieved results...")
+            for i, item in enumerate(retrieval_results):
+                query = item["query"]
+                retrieved_docs = item["retrieved_docs"]
+
+                # Prepare context from retrieved docs
+                context = "\n\n".join([doc["text"] for doc in retrieved_docs])
+                prompt = create_prompt(context, query, "emails")
+
+                # Time generation only
+                gen_start = time.time()
+                response = llm_func(prompt)
+                gen_time = time.time() - gen_start
+
+                generation_times.append(gen_time)
+                responses.append(response)
+
+                if i < 3:
+                    print(f"  Q{i + 1}: Gen={gen_time:.3f}s")
+
+            avg_gen_time = sum(generation_times) / len(generation_times)
+
+            print("\n📊 Generation Results:")
+            print(f"  Total Queries: {len(retrieval_results)}")
+            print(f"  Avg Generation Time: {avg_gen_time:.3f}s")
+            print("  (Search time from Stage 4)")
+
+            results_out["stage5"] = {
+                "total_queries": len(retrieval_results),
+                "avg_generation_time": avg_gen_time,
+                "generation_times": generation_times,
+                "responses": responses,
+            }
+
+            # Show sample results
+            print("\n📝 Sample Results:")
+            for i in range(min(3, len(retrieval_results))):
+                query = retrieval_results[i]["query"]
+                response = responses[i]
+                print(f"  Q{i + 1}: {query[:60]}...")
+                print(f"  A{i + 1}: {response[:100]}...")
+                print()
+
+        except Exception as e:
+            print(f"❌ Generation evaluation failed: {e}")
+            print("💡 Make sure transformers/vllm is installed and model is available")
+
+        print("✅ Stage 5 completed!\n")
+
+    if args.output and results_out:
+        with open(args.output, "w", encoding="utf-8") as f:
+            json.dump(results_out, f, indent=2)
+        print(f"📝 Saved results to {args.output}")
+
+
+if __name__ == "__main__":
+    main()

benchmarks/enron_emails/setup_enron_emails.py

@@ -0,0 +1,359 @@
+"""
+Enron Emails Benchmark Setup Script
+Prepares passages from emails.csv, builds LEANN index, and FAISS Flat baseline
+"""
+
+import argparse
+import csv
+import json
+import os
+import re
+from collections.abc import Iterable
+from email import message_from_string
+from email.policy import default
+from pathlib import Path
+from typing import Optional
+
+from leann import LeannBuilder
+
+
+class EnronSetup:
+    def __init__(self, data_dir: str = "data"):
+        self.data_dir = Path(data_dir)
+        self.data_dir.mkdir(parents=True, exist_ok=True)
+
+        self.passages_preview = self.data_dir / "enron_passages_preview.jsonl"
+        self.index_path = self.data_dir / "enron_index_hnsw.leann"
+        self.queries_file = self.data_dir / "evaluation_queries.jsonl"
+        self.downloads_dir = self.data_dir / "downloads"
+        self.downloads_dir.mkdir(parents=True, exist_ok=True)
+
+    # ----------------------------
+    # Dataset acquisition
+    # ----------------------------
+    def ensure_emails_csv(self, emails_csv: Optional[str]) -> str:
+        """Return a path to emails.csv, downloading from Kaggle if needed."""
+        if emails_csv:
+            p = Path(emails_csv)
+            if not p.exists():
+                raise FileNotFoundError(f"emails.csv not found: {emails_csv}")
+            return str(p)
+
+        print(
+            "📥 Trying to download Enron emails.csv from Kaggle (wcukierski/enron-email-dataset)..."
+        )
+        try:
+            from kaggle.api.kaggle_api_extended import KaggleApi
+
+            api = KaggleApi()
+            api.authenticate()
+            api.dataset_download_files(
+                "wcukierski/enron-email-dataset", path=str(self.downloads_dir), unzip=True
+            )
+            candidate = self.downloads_dir / "emails.csv"
+            if candidate.exists():
+                print(f"✅ Downloaded emails.csv: {candidate}")
+                return str(candidate)
+            else:
+                raise FileNotFoundError(
+                    f"emails.csv was not found in {self.downloads_dir} after Kaggle download"
+                )
+        except Exception as e:
+            print(
+                "❌ Could not download via Kaggle automatically. Provide --emails-csv or configure Kaggle API."
+            )
+            print(
+                "   Set KAGGLE_USERNAME and KAGGLE_KEY env vars, or place emails.csv locally and pass --emails-csv."
+            )
+            raise e
+
+    # ----------------------------
+    # Data preparation
+    # ----------------------------
+    @staticmethod
+    def _extract_message_id(raw_email: str) -> str:
+        msg = message_from_string(raw_email, policy=default)
+        val = msg.get("Message-ID", "")
+        if val.startswith("<") and val.endswith(">"):
+            val = val[1:-1]
+        return val or ""
+
+    @staticmethod
+    def _split_header_body(raw_email: str) -> tuple[str, str]:
+        parts = raw_email.split("\n\n", 1)
+        if len(parts) == 2:
+            return parts[0].strip(), parts[1].strip()
+        # Heuristic fallback
+        first_lines = raw_email.splitlines()
+        if first_lines and ":" in first_lines[0]:
+            return raw_email.strip(), ""
+        return "", raw_email.strip()
+
+    @staticmethod
+    def _split_fixed_words(text: str, chunk_words: int, keep_last: bool) -> list[str]:
+        text = (text or "").strip()
+        if not text:
+            return []
+        if chunk_words <= 0:
+            return [text]
+        words = text.split()
+        if not words:
+            return []
+        limit = len(words)
+        if not keep_last:
+            limit = (len(words) // chunk_words) * chunk_words
+        if limit == 0:
+            return []
+        chunks = [" ".join(words[i : i + chunk_words]) for i in range(0, limit, chunk_words)]
+        return [c for c in (s.strip() for s in chunks) if c]
+
+    def _iter_passages_from_csv(
+        self,
+        emails_csv: Path,
+        chunk_words: int = 256,
+        keep_last_header: bool = True,
+        keep_last_body: bool = True,
+        max_emails: int | None = None,
+    ) -> Iterable[dict]:
+        with open(emails_csv, encoding="utf-8") as f:
+            reader = csv.DictReader(f)
+            count = 0
+            for i, row in enumerate(reader):
+                if max_emails is not None and count >= max_emails:
+                    break
+
+                raw_message = row.get("message", "")
+                email_file_id = row.get("file", "")
+
+                if not raw_message.strip():
+                    continue
+
+                message_id = self._extract_message_id(raw_message)
+                if not message_id:
+                    # Fallback ID based on CSV position and file path
+                    safe_file = re.sub(r"[^A-Za-z0-9_.-]", "_", email_file_id)
+                    message_id = f"enron_{i}_{safe_file}"
+
+                header, body = self._split_header_body(raw_message)
+
+                # Header chunks
+                for chunk in self._split_fixed_words(header, chunk_words, keep_last_header):
+                    yield {
+                        "text": chunk,
+                        "metadata": {
+                            "message_id": message_id,
+                            "is_header": True,
+                            "email_file_id": email_file_id,
+                        },
+                    }
+
+                # Body chunks
+                for chunk in self._split_fixed_words(body, chunk_words, keep_last_body):
+                    yield {
+                        "text": chunk,
+                        "metadata": {
+                            "message_id": message_id,
+                            "is_header": False,
+                            "email_file_id": email_file_id,
+                        },
+                    }
+
+                count += 1
+
+    # ----------------------------
+    # Build LEANN index and FAISS baseline
+    # ----------------------------
+    def build_leann_index(
+        self,
+        emails_csv: Optional[str],
+        backend: str = "hnsw",
+        embedding_model: str = "sentence-transformers/all-mpnet-base-v2",
+        chunk_words: int = 256,
+        max_emails: int | None = None,
+    ) -> str:
+        emails_csv_path = self.ensure_emails_csv(emails_csv)
+        print(f"🏗️ Building LEANN index from {emails_csv_path}...")
+
+        builder = LeannBuilder(
+            backend_name=backend,
+            embedding_model=embedding_model,
+            embedding_mode="sentence-transformers",
+            graph_degree=32,
+            complexity=64,
+            is_recompute=True,
+            is_compact=True,
+            num_threads=4,
+        )
+
+        # Stream passages and add to builder
+        preview_written = 0
+        with open(self.passages_preview, "w", encoding="utf-8") as preview_out:
+            for p in self._iter_passages_from_csv(
+                Path(emails_csv_path), chunk_words=chunk_words, max_emails=max_emails
+            ):
+                builder.add_text(p["text"], metadata=p["metadata"])
+                if preview_written < 200:
+                    preview_out.write(json.dumps({"text": p["text"][:200], **p["metadata"]}) + "\n")
+                    preview_written += 1
+
+        print(f"🔨 Building index at {self.index_path}...")
+        builder.build_index(str(self.index_path))
+        print("✅ LEANN index built!")
+        return str(self.index_path)
+
+    def build_faiss_flat_baseline(self, index_path: str, output_dir: str = "baseline") -> str:
+        print("🔨 Building FAISS Flat baseline from LEANN passages...")
+
+        import pickle
+
+        import numpy as np
+        from leann.api import compute_embeddings
+        from leann_backend_hnsw import faiss
+
+        os.makedirs(output_dir, exist_ok=True)
+        baseline_path = os.path.join(output_dir, "faiss_flat.index")
+        metadata_path = os.path.join(output_dir, "metadata.pkl")
+
+        if os.path.exists(baseline_path) and os.path.exists(metadata_path):
+            print(f"✅ Baseline already exists at {baseline_path}")
+            return baseline_path
+
+        # Read meta for passage source and embedding model
+        meta_path = f"{index_path}.meta.json"
+        with open(meta_path, encoding="utf-8") as f:
+            meta = json.load(f)
+
+        embedding_model = meta["embedding_model"]
+        passage_source = meta["passage_sources"][0]
+        passage_file = passage_source["path"]
+
+        if not os.path.isabs(passage_file):
+            index_dir = os.path.dirname(index_path)
+            passage_file = os.path.join(index_dir, os.path.basename(passage_file))
+
+        # Load passages from builder output so IDs match LEANN
+        passages: list[str] = []
+        passage_ids: list[str] = []
+        with open(passage_file, encoding="utf-8") as f:
+            for line in f:
+                if not line.strip():
+                    continue
+                data = json.loads(line)
+                passages.append(data["text"])
+                passage_ids.append(data["id"])  # builder-assigned ID
+
+        print(f"📄 Loaded {len(passages)} passages for baseline")
+        print(f"🤖 Embedding model: {embedding_model}")
+
+        embeddings = compute_embeddings(
+            passages,
+            embedding_model,
+            mode="sentence-transformers",
+            use_server=False,
+        )
+
+        # Build FAISS IndexFlatIP
+        dim = embeddings.shape[1]
+        index = faiss.IndexFlatIP(dim)
+        emb_f32 = embeddings.astype(np.float32)
+        index.add(emb_f32.shape[0], faiss.swig_ptr(emb_f32))
+
+        faiss.write_index(index, baseline_path)
+        with open(metadata_path, "wb") as pf:
+            pickle.dump(passage_ids, pf)
+
+        print(f"✅ FAISS baseline saved: {baseline_path}")
+        print(f"✅ Metadata saved: {metadata_path}")
+        print(f"📊 Total vectors: {index.ntotal}")
+        return baseline_path
+
+    # ----------------------------
+    # Queries (optional): prepare evaluation queries file
+    # ----------------------------
+    def prepare_queries(self, min_realism: float = 0.85) -> Path:
+        print(
+            "📝 Preparing evaluation queries from HuggingFace dataset corbt/enron_emails_sample_questions ..."
+        )
+        try:
+            from datasets import load_dataset
+
+            ds = load_dataset("corbt/enron_emails_sample_questions", split="train")
+        except Exception as e:
+            print(f"⚠️  Failed to load dataset: {e}")
+            return self.queries_file
+
+        kept = 0
+        with open(self.queries_file, "w", encoding="utf-8") as out:
+            for i, item in enumerate(ds):
+                how_realistic = float(item.get("how_realistic", 0.0))
+                if how_realistic < min_realism:
+                    continue
+                qid = str(item.get("id", f"enron_q_{i}"))
+                query = item.get("question", "")
+                if not query:
+                    continue
+                record = {
+                    "id": qid,
+                    "query": query,
+                    # For reference only, not used in recall metric below
+                    "gt_message_ids": item.get("message_ids", []),
+                }
+                out.write(json.dumps(record) + "\n")
+                kept += 1
+        print(f"✅ Wrote {kept} queries to {self.queries_file}")
+        return self.queries_file
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Setup Enron Emails Benchmark")
+    parser.add_argument(
+        "--emails-csv",
+        help="Path to emails.csv (Enron dataset). If omitted, attempt Kaggle download.",
+    )
+    parser.add_argument("--data-dir", default="data", help="Data directory")
+    parser.add_argument("--backend", choices=["hnsw", "diskann"], default="hnsw")
+    parser.add_argument(
+        "--embedding-model",
+        default="sentence-transformers/all-mpnet-base-v2",
+        help="Embedding model for LEANN",
+    )
+    parser.add_argument("--chunk-words", type=int, default=256, help="Fixed word chunk size")
+    parser.add_argument("--max-emails", type=int, help="Limit number of emails to process")
+    parser.add_argument("--skip-queries", action="store_true", help="Skip creating queries file")
+    parser.add_argument("--skip-build", action="store_true", help="Skip building LEANN index")
+
+    args = parser.parse_args()
+
+    setup = EnronSetup(args.data_dir)
+
+    # Build index
+    if not args.skip_build:
+        index_path = setup.build_leann_index(
+            emails_csv=args.emails_csv,
+            backend=args.backend,
+            embedding_model=args.embedding_model,
+            chunk_words=args.chunk_words,
+            max_emails=args.max_emails,
+        )
+
+        # Build FAISS baseline from the same passages & embeddings
+        setup.build_faiss_flat_baseline(index_path)
+    else:
+        print("⏭️  Skipping LEANN index build and baseline")
+
+    # Queries file (optional)
+    if not args.skip_queries:
+        setup.prepare_queries()
+    else:
+        print("⏭️  Skipping query preparation")
+
+    print("\n🎉 Enron Emails setup completed!")
+    print(f"📁 Data directory: {setup.data_dir.absolute()}")
+    print("Next steps:")
+    print(
+        "1) Evaluate recall: python evaluate_enron_emails.py --index data/enron_index_hnsw.leann --stage 2"
+    )
+
+
+if __name__ == "__main__":
+    main()

benchmarks/financebench/README.md

@@ -0,0 +1,115 @@
+# FinanceBench Benchmark for LEANN-RAG
+
+FinanceBench is a benchmark for evaluating retrieval-augmented generation (RAG) systems on financial document question-answering tasks.
+
+## Dataset
+
+- **Source**: [PatronusAI/financebench](https://huggingface.co/datasets/PatronusAI/financebench)
+- **Questions**: 150 financial Q&A examples
+- **Documents**: 368 PDF files (10-K, 10-Q, 8-K, earnings reports)
+- **Companies**: Major public companies (3M, Apple, Microsoft, Amazon, etc.)
+- **Paper**: [FinanceBench: A New Benchmark for Financial Question Answering](https://arxiv.org/abs/2311.11944)
+
+## Structure
+
+```
+benchmarks/financebench/
+├── setup_financebench.py        # Downloads PDFs and builds index
+├── evaluate_financebench.py     # Intelligent evaluation script
+├── data/
+│   ├── financebench_merged.jsonl     # Q&A dataset
+│   ├── pdfs/                         # Downloaded financial documents
+│   └── index/                        # LEANN indexes
+│       └── financebench_full_hnsw.leann
+└── README.md
+```
+
+## Usage
+
+### 1. Setup (Download & Build Index)
+
+```bash
+cd benchmarks/financebench
+python setup_financebench.py
+```
+
+This will:
+- Download the 150 Q&A examples
+- Download all 368 PDF documents (parallel processing)
+- Build a LEANN index from 53K+ text chunks
+- Verify setup with test query
+
+### 2. Evaluation
+
+```bash
+# Basic retrieval evaluation
+python evaluate_financebench.py --index data/index/financebench_full_hnsw.leann
+
+
+# RAG generation evaluation with Qwen3-8B
+python evaluate_financebench.py --index data/index/financebench_full_hnsw.leann --stage 4 --complexity 64 --llm-backend hf --model-name Qwen/Qwen3-8B --output results_qwen3.json
+```
+
+## Evaluation Methods
+
+### Retrieval Evaluation
+Uses intelligent matching with three strategies:
+1. **Exact text overlap** - Direct substring matches
+2. **Number matching** - Key financial figures ($1,577, 1.2B, etc.)
+3. **Semantic similarity** - Word overlap with 20% threshold
+
+### QA Evaluation
+LLM-based answer evaluation using GPT-4o:
+- Handles numerical rounding and equivalent representations
+- Considers fractions, percentages, and decimal equivalents
+- Evaluates semantic meaning rather than exact text match
+
+## Benchmark Results
+
+### LEANN-RAG Performance (sentence-transformers/all-mpnet-base-v2)
+
+**Retrieval Metrics:**
+- **Question Coverage**: 100.0% (all questions retrieve relevant docs)
+- **Exact Match Rate**: 0.7% (substring overlap with evidence)
+- **Number Match Rate**: 120.7% (key financial figures matched)*
+- **Semantic Match Rate**: 4.7% (word overlap ≥20%)
+- **Average Search Time**: 0.097s
+
+**QA Metrics:**
+- **Accuracy**: 42.7% (LLM-evaluated answer correctness)
+- **Average QA Time**: 4.71s (end-to-end response time)
+
+**System Performance:**
+- **Index Size**: 53,985 chunks from 368 PDFs
+- **Build Time**: ~5-10 minutes with sentence-transformers/all-mpnet-base-v2
+
+*Note: Number match rate >100% indicates multiple retrieved documents contain the same financial figures, which is expected behavior for financial data appearing across multiple document sections.
+
+### LEANN-RAG Generation Performance (Qwen3-8B)
+
+- **Stage 4 (Index Comparison):**
+  - Compact Index: 5.0 MB
+  - Non-compact Index: 172.2 MB
+  - **Storage Saving**: 97.1%
+- **Search Performance**:
+  - Non-compact (no recompute): 0.009s avg per query
+  - Compact (with recompute): 2.203s avg per query
+  - Speed ratio: 0.004x
+
+**Generation Evaluation (20 queries, complexity=64):**
+- **Average Search Time**: 1.638s per query
+- **Average Generation Time**: 45.957s per query
+- **LLM Backend**: HuggingFace transformers
+- **Model**: Qwen/Qwen3-8B (thinking model with <think></think> processing)
+- **Total Questions Processed**: 20
+
+## Options
+
+```bash
+# Use different backends
+python setup_financebench.py --backend diskann
+python evaluate_financebench.py --index data/index/financebench_full_diskann.leann
+
+# Use different embedding models
+python setup_financebench.py --embedding-model facebook/contriever
+```

benchmarks/financebench/evaluate_financebench.py

@@ -0,0 +1,923 @@
+"""
+FinanceBench Evaluation Script - Modular Recall-based Evaluation
+"""
+
+import argparse
+import json
+import logging
+import os
+import pickle
+import time
+from pathlib import Path
+from typing import Optional
+
+import numpy as np
+import openai
+from leann import LeannChat, LeannSearcher
+from leann_backend_hnsw import faiss
+
+from ..llm_utils import evaluate_rag, generate_hf, generate_vllm, load_hf_model, load_vllm_model
+
+# Setup logging to reduce verbose output
+logging.basicConfig(level=logging.WARNING)
+logging.getLogger("leann.api").setLevel(logging.WARNING)
+logging.getLogger("leann_backend_hnsw").setLevel(logging.WARNING)
+
+
+class RecallEvaluator:
+    """Stage 2: Evaluate Recall@3 (searcher vs baseline)"""
+
+    def __init__(self, index_path: str, baseline_dir: str):
+        self.index_path = index_path
+        self.baseline_dir = baseline_dir
+        self.searcher = LeannSearcher(index_path)
+
+        # Load FAISS flat baseline
+        baseline_index_path = os.path.join(baseline_dir, "faiss_flat.index")
+        metadata_path = os.path.join(baseline_dir, "metadata.pkl")
+
+        self.faiss_index = faiss.read_index(baseline_index_path)
+        with open(metadata_path, "rb") as f:
+            self.passage_ids = pickle.load(f)
+        print(f"📚 Loaded FAISS flat baseline with {self.faiss_index.ntotal} vectors")
+
+    def evaluate_recall_at_3(
+        self, queries: list[str], complexity: int = 64, recompute_embeddings: bool = True
+    ) -> float:
+        """Evaluate recall@3 for given queries at specified complexity"""
+        recompute_str = "with recompute" if recompute_embeddings else "no recompute"
+        print(f"🔍 Evaluating recall@3 with complexity={complexity} ({recompute_str})...")
+
+        total_recall = 0.0
+        num_queries = len(queries)
+
+        for i, query in enumerate(queries):
+            # Get ground truth: search with FAISS flat
+            from leann.api import compute_embeddings
+
+            query_embedding = compute_embeddings(
+                [query],
+                self.searcher.embedding_model,
+                mode=self.searcher.embedding_mode,
+                use_server=False,
+            ).astype(np.float32)
+
+            # Search FAISS flat for ground truth using LEANN's modified faiss API
+            n = query_embedding.shape[0]  # Number of queries
+            k = 3  # Number of nearest neighbors
+            distances = np.zeros((n, k), dtype=np.float32)
+            labels = np.zeros((n, k), dtype=np.int64)
+
+            self.faiss_index.search(
+                n,
+                faiss.swig_ptr(query_embedding),
+                k,
+                faiss.swig_ptr(distances),
+                faiss.swig_ptr(labels),
+            )
+
+            # Extract the results
+            baseline_ids = {self.passage_ids[idx] for idx in labels[0]}
+
+            # Search with LEANN at specified complexity
+            test_results = self.searcher.search(
+                query,
+                top_k=3,
+                complexity=complexity,
+                recompute_embeddings=recompute_embeddings,
+            )
+            test_ids = {result.id for result in test_results}
+
+            # Calculate recall@3 = |intersection| / |ground_truth|
+            intersection = test_ids.intersection(baseline_ids)
+            recall = len(intersection) / 3.0  # Ground truth size is 3
+            total_recall += recall
+
+            if i < 3:  # Show first few examples
+                print(f"  Query {i + 1}: '{query[:50]}...' -> Recall@3: {recall:.3f}")
+                print(f"    FAISS ground truth: {list(baseline_ids)}")
+                print(f"    LEANN results (C={complexity}, {recompute_str}): {list(test_ids)}")
+                print(f"    Intersection: {list(intersection)}")
+
+        avg_recall = total_recall / num_queries
+        print(f"📊 Average Recall@3: {avg_recall:.3f} ({avg_recall * 100:.1f}%)")
+        return avg_recall
+
+    def cleanup(self):
+        """Cleanup resources"""
+        if hasattr(self, "searcher"):
+            self.searcher.cleanup()
+
+
+class FinanceBenchEvaluator:
+    def __init__(self, index_path: str, openai_api_key: Optional[str] = None):
+        self.index_path = index_path
+        self.openai_client = openai.OpenAI(api_key=openai_api_key) if openai_api_key else None
+
+        self.searcher = LeannSearcher(index_path)
+        self.chat = LeannChat(index_path) if openai_api_key else None
+
+    def load_dataset(self, dataset_path: str = "data/financebench_merged.jsonl"):
+        """Load FinanceBench dataset"""
+        data = []
+        with open(dataset_path, encoding="utf-8") as f:
+            for line in f:
+                if line.strip():
+                    data.append(json.loads(line))
+
+        print(f"📊 Loaded {len(data)} FinanceBench examples")
+        return data
+
+    def analyze_index_sizes(self) -> dict:
+        """Analyze index sizes with and without embeddings"""
+
+        print("📏 Analyzing index sizes...")
+
+        # Get all index-related files
+        index_path = Path(self.index_path)
+        index_dir = index_path.parent
+        index_name = index_path.stem  # Remove .leann extension
+
+        sizes = {}
+        total_with_embeddings = 0
+
+        # Core index files
+        index_file = index_dir / f"{index_name}.index"
+        meta_file = index_dir / f"{index_path.name}.meta.json"  # Keep .leann for meta file
+        passages_file = index_dir / f"{index_path.name}.passages.jsonl"  # Keep .leann for passages
+        passages_idx_file = index_dir / f"{index_path.name}.passages.idx"  # Keep .leann for idx
+
+        for file_path, name in [
+            (index_file, "index"),
+            (meta_file, "metadata"),
+            (passages_file, "passages_text"),
+            (passages_idx_file, "passages_index"),
+        ]:
+            if file_path.exists():
+                size_mb = file_path.stat().st_size / (1024 * 1024)
+                sizes[name] = size_mb
+                total_with_embeddings += size_mb
+
+            else:
+                sizes[name] = 0
+
+        sizes["total_with_embeddings"] = total_with_embeddings
+        sizes["index_only_mb"] = sizes["index"]  # Just the .index file for fair comparison
+
+        print(f"  📁 Total index size: {total_with_embeddings:.1f} MB")
+        print(f"  📁 Index file only: {sizes['index']:.1f} MB")
+
+        return sizes
+
+    def create_compact_index_for_comparison(self, compact_index_path: str) -> dict:
+        """Create a compact index for comparison purposes"""
+        print("🏗️ Building compact index from existing passages...")
+
+        # Load existing passages from current index
+
+        from leann import LeannBuilder
+
+        current_index_path = Path(self.index_path)
+        current_index_dir = current_index_path.parent
+        current_index_name = current_index_path.name
+
+        # Read metadata to get passage source
+        meta_path = current_index_dir / f"{current_index_name}.meta.json"
+        with open(meta_path) as f:
+            import json
+
+            meta = json.load(f)
+
+        passage_source = meta["passage_sources"][0]
+        passage_file = passage_source["path"]
+
+        # Convert relative path to absolute
+        if not Path(passage_file).is_absolute():
+            passage_file = current_index_dir / Path(passage_file).name
+
+        print(f"📄 Loading passages from {passage_file}...")
+
+        # Build compact index with same passages
+        builder = LeannBuilder(
+            backend_name="hnsw",
+            embedding_model=meta["embedding_model"],
+            embedding_mode=meta.get("embedding_mode", "sentence-transformers"),
+            is_recompute=True,  # Enable recompute (no stored embeddings)
+            is_compact=True,  # Enable compact storage
+            **meta.get("backend_kwargs", {}),
+        )
+
+        # Load all passages
+        with open(passage_file, encoding="utf-8") as f:
+            for line in f:
+                if line.strip():
+                    data = json.loads(line)
+                    builder.add_text(data["text"], metadata=data.get("metadata", {}))
+
+        print(f"🔨 Building compact index at {compact_index_path}...")
+        builder.build_index(compact_index_path)
+
+        # Analyze the compact index size
+        temp_evaluator = FinanceBenchEvaluator(compact_index_path)
+        compact_sizes = temp_evaluator.analyze_index_sizes()
+        compact_sizes["index_type"] = "compact"
+
+        return compact_sizes
+
+    def create_non_compact_index_for_comparison(self, non_compact_index_path: str) -> dict:
+        """Create a non-compact index for comparison purposes"""
+        print("🏗️ Building non-compact index from existing passages...")
+
+        # Load existing passages from current index
+
+        from leann import LeannBuilder
+
+        current_index_path = Path(self.index_path)
+        current_index_dir = current_index_path.parent
+        current_index_name = current_index_path.name
+
+        # Read metadata to get passage source
+        meta_path = current_index_dir / f"{current_index_name}.meta.json"
+        with open(meta_path) as f:
+            import json
+
+            meta = json.load(f)
+
+        passage_source = meta["passage_sources"][0]
+        passage_file = passage_source["path"]
+
+        # Convert relative path to absolute
+        if not Path(passage_file).is_absolute():
+            passage_file = current_index_dir / Path(passage_file).name
+
+        print(f"📄 Loading passages from {passage_file}...")
+
+        # Build non-compact index with same passages
+        builder = LeannBuilder(
+            backend_name="hnsw",
+            embedding_model=meta["embedding_model"],
+            embedding_mode=meta.get("embedding_mode", "sentence-transformers"),
+            is_recompute=False,  # Disable recompute (store embeddings)
+            is_compact=False,  # Disable compact storage
+            **{
+                k: v
+                for k, v in meta.get("backend_kwargs", {}).items()
+                if k not in ["is_recompute", "is_compact"]
+            },
+        )
+
+        # Load all passages
+        with open(passage_file, encoding="utf-8") as f:
+            for line in f:
+                if line.strip():
+                    data = json.loads(line)
+                    builder.add_text(data["text"], metadata=data.get("metadata", {}))
+
+        print(f"🔨 Building non-compact index at {non_compact_index_path}...")
+        builder.build_index(non_compact_index_path)
+
+        # Analyze the non-compact index size
+        temp_evaluator = FinanceBenchEvaluator(non_compact_index_path)
+        non_compact_sizes = temp_evaluator.analyze_index_sizes()
+        non_compact_sizes["index_type"] = "non_compact"
+
+        return non_compact_sizes
+
+    def compare_index_performance(
+        self, non_compact_path: str, compact_path: str, test_data: list, complexity: int
+    ) -> dict:
+        """Compare performance between non-compact and compact indexes"""
+        print("⚡ Comparing search performance between indexes...")
+
+        import time
+
+        from leann import LeannSearcher
+
+        # Test queries
+        test_queries = [item["question"] for item in test_data[:5]]
+
+        results = {
+            "non_compact": {"search_times": []},
+            "compact": {"search_times": []},
+            "avg_search_times": {},
+            "speed_ratio": 0.0,
+        }
+
+        # Test non-compact index (no recompute)
+        print("  🔍 Testing non-compact index (no recompute)...")
+        non_compact_searcher = LeannSearcher(non_compact_path)
+
+        for query in test_queries:
+            start_time = time.time()
+            _ = non_compact_searcher.search(
+                query, top_k=3, complexity=complexity, recompute_embeddings=False
+            )
+            search_time = time.time() - start_time
+            results["non_compact"]["search_times"].append(search_time)
+
+        # Test compact index (with recompute)
+        print("  🔍 Testing compact index (with recompute)...")
+        compact_searcher = LeannSearcher(compact_path)
+
+        for query in test_queries:
+            start_time = time.time()
+            _ = compact_searcher.search(
+                query, top_k=3, complexity=complexity, recompute_embeddings=True
+            )
+            search_time = time.time() - start_time
+            results["compact"]["search_times"].append(search_time)
+
+        # Calculate averages
+        results["avg_search_times"]["non_compact"] = sum(
+            results["non_compact"]["search_times"]
+        ) / len(results["non_compact"]["search_times"])
+        results["avg_search_times"]["compact"] = sum(results["compact"]["search_times"]) / len(
+            results["compact"]["search_times"]
+        )
+
+        # Performance ratio
+        if results["avg_search_times"]["compact"] > 0:
+            results["speed_ratio"] = (
+                results["avg_search_times"]["non_compact"] / results["avg_search_times"]["compact"]
+            )
+        else:
+            results["speed_ratio"] = float("inf")
+
+        print(
+            f"    Non-compact (no recompute): {results['avg_search_times']['non_compact']:.3f}s avg"
+        )
+        print(f"    Compact (with recompute): {results['avg_search_times']['compact']:.3f}s avg")
+        print(f"    Speed ratio: {results['speed_ratio']:.2f}x")
+
+        # Cleanup
+        non_compact_searcher.cleanup()
+        compact_searcher.cleanup()
+
+        return results
+
+    def evaluate_timing_breakdown(
+        self, data: list[dict], max_samples: Optional[int] = None
+    ) -> dict:
+        """Evaluate timing breakdown and accuracy by hacking LeannChat.ask() for separated timing"""
+        if not self.chat or not self.openai_client:
+            print("⚠️  Skipping timing evaluation (no OpenAI API key provided)")
+            return {
+                "total_questions": 0,
+                "avg_search_time": 0.0,
+                "avg_generation_time": 0.0,
+                "avg_total_time": 0.0,
+                "accuracy": 0.0,
+            }
+
+        print("🔍🤖 Evaluating timing breakdown and accuracy (search + generation)...")
+
+        if max_samples:
+            data = data[:max_samples]
+            print(f"📝 Using first {max_samples} samples for timing evaluation")
+
+        search_times = []
+        generation_times = []
+        total_times = []
+        correct_answers = 0
+
+        for i, item in enumerate(data):
+            question = item["question"]
+            ground_truth = item["answer"]
+
+            try:
+                # Hack: Monkey-patch the ask method to capture internal timing
+                original_ask = self.chat.ask
+                captured_search_time = None
+                captured_generation_time = None
+
+                def patched_ask(*args, **kwargs):
+                    nonlocal captured_search_time, captured_generation_time
+
+                    # Time the search part
+                    search_start = time.time()
+                    results = self.chat.searcher.search(args[0], top_k=3, complexity=64)
+                    captured_search_time = time.time() - search_start
+
+                    # Time the generation part
+                    context = "\n\n".join([r.text for r in results])
+                    prompt = (
+                        "Here is some retrieved context that might help answer your question:\n\n"
+                        f"{context}\n\n"
+                        f"Question: {args[0]}\n\n"
+                        "Please provide the best answer you can based on this context and your knowledge."
+                    )
+
+                    generation_start = time.time()
+                    answer = self.chat.llm.ask(prompt)
+                    captured_generation_time = time.time() - generation_start
+
+                    return answer
+
+                # Apply the patch
+                self.chat.ask = patched_ask
+
+                # Time the total QA
+                total_start = time.time()
+                generated_answer = self.chat.ask(question)
+                total_time = time.time() - total_start
+
+                # Restore original method
+                self.chat.ask = original_ask
+
+                # Store the timings
+                search_times.append(captured_search_time)
+                generation_times.append(captured_generation_time)
+                total_times.append(total_time)
+
+                # Check accuracy using LLM as judge
+                is_correct = self._check_answer_accuracy(generated_answer, ground_truth, question)
+                if is_correct:
+                    correct_answers += 1
+
+                status = "✅" if is_correct else "❌"
+                print(
+                    f"Question {i + 1}/{len(data)}: {status} Search={captured_search_time:.3f}s, Gen={captured_generation_time:.3f}s, Total={total_time:.3f}s"
+                )
+                print(f"  GT: {ground_truth}")
+                print(f"  Gen: {generated_answer[:100]}...")
+
+            except Exception as e:
+                print(f"  ❌ Error: {e}")
+                search_times.append(0.0)
+                generation_times.append(0.0)
+                total_times.append(0.0)
+
+        accuracy = correct_answers / len(data) if data else 0.0
+
+        metrics = {
+            "total_questions": len(data),
+            "avg_search_time": sum(search_times) / len(search_times) if search_times else 0.0,
+            "avg_generation_time": sum(generation_times) / len(generation_times)
+            if generation_times
+            else 0.0,
+            "avg_total_time": sum(total_times) / len(total_times) if total_times else 0.0,
+            "accuracy": accuracy,
+            "correct_answers": correct_answers,
+            "search_times": search_times,
+            "generation_times": generation_times,
+            "total_times": total_times,
+        }
+
+        return metrics
+
+    def _check_answer_accuracy(
+        self, generated_answer: str, ground_truth: str, question: str
+    ) -> bool:
+        """Check if generated answer matches ground truth using LLM as judge"""
+        judge_prompt = f"""You are an expert judge evaluating financial question answering.
+
+Question: {question}
+
+Ground Truth Answer: {ground_truth}
+
+Generated Answer: {generated_answer}
+
+Task: Determine if the generated answer is factually correct compared to the ground truth. Focus on:
+1. Numerical accuracy (exact values, units, currency)
+2. Key financial concepts and terminology
+3. Overall factual correctness
+
+For financial data, small formatting differences are OK (e.g., "$1,577" vs "1577 million" vs "$1.577 billion"), but the core numerical value must match.
+
+Respond with exactly one word: "CORRECT" if the generated answer is factually accurate, or "INCORRECT" if it's wrong or significantly different."""
+
+        try:
+            judge_response = self.openai_client.chat.completions.create(
+                model="gpt-4o-mini",
+                messages=[{"role": "user", "content": judge_prompt}],
+                max_tokens=10,
+                temperature=0,
+            )
+            judgment = judge_response.choices[0].message.content.strip().upper()
+            return judgment == "CORRECT"
+        except Exception as e:
+            print(f"  ⚠️  Judge error: {e}, falling back to string matching")
+            # Fallback to simple string matching
+            gen_clean = generated_answer.strip().lower().replace("$", "").replace(",", "")
+            gt_clean = ground_truth.strip().lower().replace("$", "").replace(",", "")
+            return gt_clean in gen_clean
+
+    def _print_results(self, timing_metrics: dict):
+        """Print evaluation results"""
+        print("\n🎯 EVALUATION RESULTS")
+        print("=" * 50)
+
+        # Index comparison analysis
+        if "current_index" in timing_metrics and "non_compact_index" in timing_metrics:
+            print("\n📏 Index Comparison Analysis:")
+            current = timing_metrics["current_index"]
+            non_compact = timing_metrics["non_compact_index"]
+
+            print(f"  Compact index (current): {current.get('total_with_embeddings', 0):.1f} MB")
+            print(
+                f"  Non-compact index (with embeddings): {non_compact.get('total_with_embeddings', 0):.1f} MB"
+            )
+            print(
+                f"  Storage saving by compact: {timing_metrics.get('storage_saving_percent', 0):.1f}%"
+            )
+
+            print("  Component breakdown (non-compact):")
+            print(f"    - Main index: {non_compact.get('index', 0):.1f} MB")
+            print(f"    - Passages text: {non_compact.get('passages_text', 0):.1f} MB")
+            print(f"    - Passages index: {non_compact.get('passages_index', 0):.1f} MB")
+            print(f"    - Metadata: {non_compact.get('metadata', 0):.1f} MB")
+
+        # Performance comparison
+        if "performance_comparison" in timing_metrics:
+            perf = timing_metrics["performance_comparison"]
+            print("\n⚡ Performance Comparison:")
+            print(
+                f"  Non-compact (no recompute): {perf.get('avg_search_times', {}).get('non_compact', 0):.3f}s avg"
+            )
+            print(
+                f"  Compact (with recompute): {perf.get('avg_search_times', {}).get('compact', 0):.3f}s avg"
+            )
+            print(f"  Speed ratio: {perf.get('speed_ratio', 0):.2f}x")
+
+        # Legacy single index analysis (fallback)
+        if "total_with_embeddings" in timing_metrics and "current_index" not in timing_metrics:
+            print("\n📏 Index Size Analysis:")
+            print(f"  Total index size: {timing_metrics.get('total_with_embeddings', 0):.1f} MB")
+
+        print("\n📊 Accuracy:")
+        print(f"  Accuracy: {timing_metrics.get('accuracy', 0) * 100:.1f}%")
+        print(
+            f"  Correct Answers: {timing_metrics.get('correct_answers', 0)}/{timing_metrics.get('total_questions', 0)}"
+        )
+
+        print("\n📊 Timing Breakdown:")
+        print(f"  Total Questions: {timing_metrics.get('total_questions', 0)}")
+        print(f"  Avg Search Time: {timing_metrics.get('avg_search_time', 0):.3f}s")
+        print(f"  Avg Generation Time: {timing_metrics.get('avg_generation_time', 0):.3f}s")
+        print(f"  Avg Total Time: {timing_metrics.get('avg_total_time', 0):.3f}s")
+
+        if timing_metrics.get("avg_total_time", 0) > 0:
+            search_pct = (
+                timing_metrics.get("avg_search_time", 0)
+                / timing_metrics.get("avg_total_time", 1)
+                * 100
+            )
+            gen_pct = (
+                timing_metrics.get("avg_generation_time", 0)
+                / timing_metrics.get("avg_total_time", 1)
+                * 100
+            )
+            print("\n📈 Time Distribution:")
+            print(f"  Search: {search_pct:.1f}%")
+            print(f"  Generation: {gen_pct:.1f}%")
+
+    def cleanup(self):
+        """Cleanup resources"""
+        if self.searcher:
+            self.searcher.cleanup()
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Modular FinanceBench Evaluation")
+    parser.add_argument("--index", required=True, help="Path to LEANN index")
+    parser.add_argument("--dataset", default="data/financebench_merged.jsonl", help="Dataset path")
+    parser.add_argument(
+        "--stage",
+        choices=["2", "3", "4", "all"],
+        default="all",
+        help="Which stage to run (2=recall, 3=complexity, 4=generation)",
+    )
+    parser.add_argument("--complexity", type=int, default=None, help="Complexity for search")
+    parser.add_argument("--baseline-dir", default="baseline", help="Baseline output directory")
+    parser.add_argument("--openai-api-key", help="OpenAI API key for generation evaluation")
+    parser.add_argument("--output", help="Save results to JSON file")
+    parser.add_argument(
+        "--llm-backend", choices=["openai", "hf", "vllm"], default="openai", help="LLM backend"
+    )
+    parser.add_argument("--model-name", default="Qwen3-8B", help="Model name for HF/vLLM")
+
+    args = parser.parse_args()
+
+    try:
+        # Check if baseline exists
+        baseline_index_path = os.path.join(args.baseline_dir, "faiss_flat.index")
+        if not os.path.exists(baseline_index_path):
+            print(f"❌ FAISS baseline not found at {baseline_index_path}")
+            print("💡 Please run setup_financebench.py first to build the baseline")
+            exit(1)
+
+        if args.stage == "2" or args.stage == "all":
+            # Stage 2: Recall@3 evaluation
+            print("🚀 Starting Stage 2: Recall@3 evaluation")
+
+            evaluator = RecallEvaluator(args.index, args.baseline_dir)
+
+            # Load FinanceBench queries for testing
+            print("📖 Loading FinanceBench dataset...")
+            queries = []
+            with open(args.dataset, encoding="utf-8") as f:
+                for line in f:
+                    if line.strip():
+                        data = json.loads(line)
+                        queries.append(data["question"])
+
+            # Test with more queries for robust measurement
+            test_queries = queries[:2000]
+            print(f"🧪 Testing with {len(test_queries)} queries")
+
+            # Test with complexity 64
+            complexity = 64
+            recall = evaluator.evaluate_recall_at_3(test_queries, complexity)
+            print(f"📈 Recall@3 at complexity {complexity}: {recall * 100:.1f}%")
+
+            evaluator.cleanup()
+            print("✅ Stage 2 completed!\n")
+
+        # Shared non-compact index path for Stage 3 and 4
+        non_compact_index_path = args.index.replace(".leann", "_noncompact.leann")
+        complexity = args.complexity
+
+        if args.stage == "3" or args.stage == "all":
+            # Stage 3: Binary search for 90% recall complexity (using non-compact index for speed)
+            print("🚀 Starting Stage 3: Binary search for 90% recall complexity")
+            print(
+                "💡 Creating non-compact index for fast binary search with recompute_embeddings=False"
+            )
+
+            # Create non-compact index for binary search (will be reused in Stage 4)
+            print("🏗️ Creating non-compact index for binary search...")
+            evaluator = FinanceBenchEvaluator(args.index)
+            evaluator.create_non_compact_index_for_comparison(non_compact_index_path)
+
+            # Use non-compact index for binary search
+            binary_search_evaluator = RecallEvaluator(non_compact_index_path, args.baseline_dir)
+
+            # Load queries for testing
+            print("📖 Loading FinanceBench dataset...")
+            queries = []
+            with open(args.dataset, encoding="utf-8") as f:
+                for line in f:
+                    if line.strip():
+                        data = json.loads(line)
+                        queries.append(data["question"])
+
+            # Use more queries for robust measurement
+            test_queries = queries[:200]
+            print(f"🧪 Testing with {len(test_queries)} queries")
+
+            # Binary search for 90% recall complexity (without recompute for speed)
+            target_recall = 0.9
+            min_complexity, max_complexity = 1, 32
+
+            print(f"🔍 Binary search for {target_recall * 100}% recall complexity...")
+            print(f"Search range: {min_complexity} to {max_complexity}")
+
+            best_complexity = None
+            best_recall = 0.0
+
+            while min_complexity <= max_complexity:
+                mid_complexity = (min_complexity + max_complexity) // 2
+
+                print(
+                    f"\n🧪 Testing complexity {mid_complexity} (no recompute, non-compact index)..."
+                )
+                # Use recompute_embeddings=False on non-compact index for fast binary search
+                recall = binary_search_evaluator.evaluate_recall_at_3(
+                    test_queries, mid_complexity, recompute_embeddings=False
+                )
+
+                print(
+                    f"  Complexity {mid_complexity}: Recall@3 = {recall:.3f} ({recall * 100:.1f}%)"
+                )
+
+                if recall >= target_recall:
+                    best_complexity = mid_complexity
+                    best_recall = recall
+                    max_complexity = mid_complexity - 1
+                    print("  ✅ Target reached! Searching for lower complexity...")
+                else:
+                    min_complexity = mid_complexity + 1
+                    print("  ❌ Below target. Searching for higher complexity...")
+
+            if best_complexity is not None:
+                print("\n🎯 Optimal complexity found!")
+                print(f"  Complexity: {best_complexity}")
+                print(f"  Recall@3: {best_recall:.3f} ({best_recall * 100:.1f}%)")
+
+                # Test a few complexities around the optimal one for verification
+                print("\n🔬 Verification test around optimal complexity:")
+                verification_complexities = [
+                    max(1, best_complexity - 2),
+                    max(1, best_complexity - 1),
+                    best_complexity,
+                    best_complexity + 1,
+                    best_complexity + 2,
+                ]
+
+                for complexity in verification_complexities:
+                    if complexity <= 512:  # reasonable upper bound
+                        recall = binary_search_evaluator.evaluate_recall_at_3(
+                            test_queries, complexity, recompute_embeddings=False
+                        )
+                        status = "✅" if recall >= target_recall else "❌"
+                        print(f"  {status} Complexity {complexity:3d}: {recall * 100:5.1f}%")
+
+                # Now test the optimal complexity with compact index and recompute for comparison
+                print(
+                    f"\n🔄 Testing optimal complexity {best_complexity} on compact index WITH recompute..."
+                )
+                compact_evaluator = RecallEvaluator(args.index, args.baseline_dir)
+                recall_with_recompute = compact_evaluator.evaluate_recall_at_3(
+                    test_queries[:10], best_complexity, recompute_embeddings=True
+                )
+                print(
+                    f"  ✅ Complexity {best_complexity} (compact index with recompute): {recall_with_recompute * 100:.1f}%"
+                )
+                complexity = best_complexity
+                print(
+                    f"  📊 Recall difference: {abs(best_recall - recall_with_recompute) * 100:.2f}%"
+                )
+                compact_evaluator.cleanup()
+            else:
+                print(f"\n❌ Could not find complexity achieving {target_recall * 100}% recall")
+                print("All tested complexities were below target.")
+
+            # Cleanup evaluators (keep non-compact index for Stage 4)
+            binary_search_evaluator.cleanup()
+            evaluator.cleanup()
+
+            print("✅ Stage 3 completed! Non-compact index saved for Stage 4.\n")
+
+        if args.stage == "4" or args.stage == "all":
+            # Stage 4: Comprehensive evaluation with dual index comparison
+            print("🚀 Starting Stage 4: Comprehensive evaluation with dual index comparison")
+
+            # Use FinanceBench evaluator for QA evaluation
+            evaluator = FinanceBenchEvaluator(
+                args.index, args.openai_api_key if args.llm_backend == "openai" else None
+            )
+
+            print("📖 Loading FinanceBench dataset...")
+            data = evaluator.load_dataset(args.dataset)
+
+            # Step 1: Analyze current (compact) index
+            print("\n📏 Analyzing current index (compact, pruned)...")
+            compact_size_metrics = evaluator.analyze_index_sizes()
+            compact_size_metrics["index_type"] = "compact"
+
+            # Step 2: Use existing non-compact index or create if needed
+            from pathlib import Path
+
+            if Path(non_compact_index_path).exists():
+                print(
+                    f"\n📁 Using existing non-compact index from Stage 3: {non_compact_index_path}"
+                )
+                temp_evaluator = FinanceBenchEvaluator(non_compact_index_path)
+                non_compact_size_metrics = temp_evaluator.analyze_index_sizes()
+                non_compact_size_metrics["index_type"] = "non_compact"
+            else:
+                print("\n🏗️ Creating non-compact index (with embeddings) for comparison...")
+                non_compact_size_metrics = evaluator.create_non_compact_index_for_comparison(
+                    non_compact_index_path
+                )
+
+            # Step 3: Compare index sizes
+            print("\n📊 Index size comparison:")
+            print(
+                f"  Compact index (current): {compact_size_metrics['total_with_embeddings']:.1f} MB"
+            )
+            print(
+                f"  Non-compact index: {non_compact_size_metrics['total_with_embeddings']:.1f} MB"
+            )
+            print("\n📊 Index-only size comparison (.index file only):")
+            print(f"  Compact index: {compact_size_metrics['index_only_mb']:.1f} MB")
+            print(f"  Non-compact index: {non_compact_size_metrics['index_only_mb']:.1f} MB")
+            # Use index-only size for fair comparison (same as Enron emails)
+            storage_saving = (
+                (non_compact_size_metrics["index_only_mb"] - compact_size_metrics["index_only_mb"])
+                / non_compact_size_metrics["index_only_mb"]
+                * 100
+            )
+            print(f"  Storage saving by compact: {storage_saving:.1f}%")
+
+            # Step 4: Performance comparison between the two indexes
+            if complexity is None:
+                raise ValueError("Complexity is required for performance comparison")
+
+            print("\n⚡ Performance comparison between indexes...")
+            performance_metrics = evaluator.compare_index_performance(
+                non_compact_index_path, args.index, data[:10], complexity=complexity
+            )
+
+            # Step 5: Generation evaluation
+            test_samples = 20
+            print(f"\n🧪 Testing with first {test_samples} samples for generation analysis")
+
+            if args.llm_backend == "openai" and args.openai_api_key:
+                print("🔍🤖 Running OpenAI-based generation evaluation...")
+                evaluation_start = time.time()
+                timing_metrics = evaluator.evaluate_timing_breakdown(data[:test_samples])
+                evaluation_time = time.time() - evaluation_start
+            else:
+                print(
+                    f"🔍🤖 Running {args.llm_backend} generation evaluation with {args.model_name}..."
+                )
+                try:
+                    # Load LLM
+                    if args.llm_backend == "hf":
+                        tokenizer, model = load_hf_model(args.model_name)
+
+                        def llm_func(prompt):
+                            return generate_hf(tokenizer, model, prompt)
+                    else:  # vllm
+                        llm, sampling_params = load_vllm_model(args.model_name)
+
+                        def llm_func(prompt):
+                            return generate_vllm(llm, sampling_params, prompt)
+
+                    # Simple generation evaluation
+                    queries = [item["question"] for item in data[:test_samples]]
+                    gen_results = evaluate_rag(
+                        evaluator.searcher,
+                        llm_func,
+                        queries,
+                        domain="finance",
+                        complexity=complexity,
+                    )
+
+                    timing_metrics = {
+                        "total_questions": len(queries),
+                        "avg_search_time": gen_results["avg_search_time"],
+                        "avg_generation_time": gen_results["avg_generation_time"],
+                        "results": gen_results["results"],
+                    }
+                    evaluation_time = time.time()
+
+                except Exception as e:
+                    print(f"❌ Generation evaluation failed: {e}")
+                    timing_metrics = {
+                        "total_questions": 0,
+                        "avg_search_time": 0,
+                        "avg_generation_time": 0,
+                    }
+                    evaluation_time = 0
+
+            # Combine all metrics
+            combined_metrics = {
+                **timing_metrics,
+                "total_evaluation_time": evaluation_time,
+                "current_index": compact_size_metrics,
+                "non_compact_index": non_compact_size_metrics,
+                "performance_comparison": performance_metrics,
+                "storage_saving_percent": storage_saving,
+            }
+
+            # Print results
+            print("\n📊 Generation Results:")
+            print(f"  Total Questions: {timing_metrics.get('total_questions', 0)}")
+            print(f"  Avg Search Time: {timing_metrics.get('avg_search_time', 0):.3f}s")
+            print(f"  Avg Generation Time: {timing_metrics.get('avg_generation_time', 0):.3f}s")
+
+            # Save results if requested
+            if args.output:
+                print(f"\n💾 Saving results to {args.output}...")
+                with open(args.output, "w") as f:
+                    json.dump(combined_metrics, f, indent=2, default=str)
+                print(f"✅ Results saved to {args.output}")
+
+            evaluator.cleanup()
+            print("✅ Stage 4 completed!\n")
+
+        if args.stage == "all":
+            print("🎉 All evaluation stages completed successfully!")
+            print("\n📋 Summary:")
+            print("  Stage 2: ✅ Recall@3 evaluation completed")
+            print("  Stage 3: ✅ Optimal complexity found")
+            print("  Stage 4: ✅ Generation accuracy & timing evaluation completed")
+            print("\n🔧 Recommended next steps:")
+            print("  - Use optimal complexity for best speed/accuracy balance")
+            print("  - Review accuracy and timing breakdown for performance optimization")
+            print("  - Run full evaluation on complete dataset if needed")
+
+            # Clean up non-compact index after all stages complete
+            print("\n🧹 Cleaning up temporary non-compact index...")
+            from pathlib import Path
+
+            if Path(non_compact_index_path).exists():
+                temp_index_dir = Path(non_compact_index_path).parent
+                temp_index_name = Path(non_compact_index_path).name
+                for temp_file in temp_index_dir.glob(f"{temp_index_name}*"):
+                    temp_file.unlink()
+                print(f"✅ Cleaned up {non_compact_index_path}")
+            else:
+                print("📝 No temporary index to clean up")
+    except KeyboardInterrupt:
+        print("\n⚠️  Evaluation interrupted by user")
+        exit(1)
+    except Exception as e:
+        print(f"\n❌ Stage {args.stage} failed: {e}")
+        exit(1)
+
+
+if __name__ == "__main__":
+    main()

benchmarks/financebench/setup_financebench.py

@@ -0,0 +1,462 @@
+#!/usr/bin/env python3
+"""
+FinanceBench Complete Setup Script
+Downloads all PDFs and builds full LEANN datastore
+"""
+
+import argparse
+import os
+import re
+import time
+from concurrent.futures import ThreadPoolExecutor, as_completed
+from pathlib import Path
+from threading import Lock
+
+import pymupdf
+import requests
+from leann import LeannBuilder, LeannSearcher
+from tqdm import tqdm
+
+
+class FinanceBenchSetup:
+    def __init__(self, data_dir: str = "data"):
+        self.base_dir = Path(__file__).parent  # benchmarks/financebench/
+        self.data_dir = self.base_dir / data_dir
+        self.pdf_dir = self.data_dir / "pdfs"
+        self.dataset_file = self.data_dir / "financebench_merged.jsonl"
+        self.index_dir = self.data_dir / "index"
+        self.download_lock = Lock()
+
+    def download_dataset(self):
+        """Download the main FinanceBench dataset"""
+        print("📊 Downloading FinanceBench dataset...")
+        self.data_dir.mkdir(parents=True, exist_ok=True)
+
+        if self.dataset_file.exists():
+            print(f"✅ Dataset already exists: {self.dataset_file}")
+            return
+
+        url = "https://huggingface.co/datasets/PatronusAI/financebench/raw/main/financebench_merged.jsonl"
+        response = requests.get(url, stream=True)
+        response.raise_for_status()
+
+        with open(self.dataset_file, "wb") as f:
+            for chunk in response.iter_content(chunk_size=8192):
+                f.write(chunk)
+
+        print(f"✅ Dataset downloaded: {self.dataset_file}")
+
+    def get_pdf_list(self):
+        """Get list of all PDF files from GitHub"""
+        print("📋 Fetching PDF list from GitHub...")
+
+        response = requests.get(
+            "https://api.github.com/repos/patronus-ai/financebench/contents/pdfs"
+        )
+        response.raise_for_status()
+        pdf_files = response.json()
+
+        print(f"Found {len(pdf_files)} PDF files")
+        return pdf_files
+
+    def download_single_pdf(self, pdf_info, position):
+        """Download a single PDF file"""
+        pdf_name = pdf_info["name"]
+        pdf_path = self.pdf_dir / pdf_name
+
+        # Skip if already downloaded
+        if pdf_path.exists() and pdf_path.stat().st_size > 0:
+            return f"✅ {pdf_name} (cached)"
+
+        try:
+            # Download PDF
+            response = requests.get(pdf_info["download_url"], timeout=60)
+            response.raise_for_status()
+
+            # Write to file
+            with self.download_lock:
+                with open(pdf_path, "wb") as f:
+                    f.write(response.content)
+
+            return f"✅ {pdf_name} ({len(response.content) // 1024}KB)"
+
+        except Exception as e:
+            return f"❌ {pdf_name}: {e!s}"
+
+    def download_all_pdfs(self, max_workers: int = 5):
+        """Download all PDF files with parallel processing"""
+        self.pdf_dir.mkdir(parents=True, exist_ok=True)
+
+        pdf_files = self.get_pdf_list()
+
+        print(f"📥 Downloading {len(pdf_files)} PDFs with {max_workers} workers...")
+
+        with ThreadPoolExecutor(max_workers=max_workers) as executor:
+            # Submit all download tasks
+            future_to_pdf = {
+                executor.submit(self.download_single_pdf, pdf_info, i): pdf_info["name"]
+                for i, pdf_info in enumerate(pdf_files)
+            }
+
+            # Process completed downloads with progress bar
+            with tqdm(total=len(pdf_files), desc="Downloading PDFs") as pbar:
+                for future in as_completed(future_to_pdf):
+                    result = future.result()
+                    pbar.set_postfix_str(result.split()[-1] if "✅" in result else "Error")
+                    pbar.update(1)
+
+        # Verify downloads
+        downloaded_pdfs = list(self.pdf_dir.glob("*.pdf"))
+        print(f"✅ Successfully downloaded {len(downloaded_pdfs)}/{len(pdf_files)} PDFs")
+
+        # Show any failures
+        missing_pdfs = []
+        for pdf_info in pdf_files:
+            pdf_path = self.pdf_dir / pdf_info["name"]
+            if not pdf_path.exists() or pdf_path.stat().st_size == 0:
+                missing_pdfs.append(pdf_info["name"])
+
+        if missing_pdfs:
+            print(f"⚠️  Failed to download {len(missing_pdfs)} PDFs:")
+            for pdf in missing_pdfs[:5]:  # Show first 5
+                print(f"   - {pdf}")
+            if len(missing_pdfs) > 5:
+                print(f"   ... and {len(missing_pdfs) - 5} more")
+
+    def build_leann_index(
+        self,
+        backend: str = "hnsw",
+        embedding_model: str = "sentence-transformers/all-mpnet-base-v2",
+    ):
+        """Build LEANN index from all PDFs"""
+        print(f"🏗️  Building LEANN index with {backend} backend...")
+
+        # Check if we have PDFs
+        pdf_files = list(self.pdf_dir.glob("*.pdf"))
+        if not pdf_files:
+            raise RuntimeError("No PDF files found! Run download first.")
+
+        print(f"Found {len(pdf_files)} PDF files to process")
+
+        start_time = time.time()
+
+        # Initialize builder with standard compact configuration
+        builder = LeannBuilder(
+            backend_name=backend,
+            embedding_model=embedding_model,
+            embedding_mode="sentence-transformers",
+            graph_degree=32,
+            complexity=64,
+            is_recompute=True,  # Enable recompute (no stored embeddings)
+            is_compact=True,  # Enable compact storage (pruned)
+            num_threads=4,
+        )
+
+        # Process PDFs and extract text
+        total_chunks = 0
+        failed_pdfs = []
+
+        for pdf_path in tqdm(pdf_files, desc="Processing PDFs"):
+            try:
+                chunks = self.extract_pdf_text(pdf_path)
+                for chunk in chunks:
+                    builder.add_text(chunk["text"], metadata=chunk["metadata"])
+                    total_chunks += 1
+
+            except Exception as e:
+                print(f"❌ Failed to process {pdf_path.name}: {e}")
+                failed_pdfs.append(pdf_path.name)
+                continue
+
+        # Build index in index directory
+        self.index_dir.mkdir(parents=True, exist_ok=True)
+        index_path = self.index_dir / f"financebench_full_{backend}.leann"
+        print(f"🔨 Building index: {index_path}")
+        builder.build_index(str(index_path))
+
+        build_time = time.time() - start_time
+
+        print("✅ Index built successfully!")
+        print(f"   📁 Index path: {index_path}")
+        print(f"   📊 Total chunks: {total_chunks:,}")
+        print(f"   📄 Processed PDFs: {len(pdf_files) - len(failed_pdfs)}/{len(pdf_files)}")
+        print(f"   ⏱️  Build time: {build_time:.1f}s")
+
+        if failed_pdfs:
+            print(f"   ⚠️  Failed PDFs: {failed_pdfs}")
+
+        return str(index_path)
+
+    def build_faiss_flat_baseline(self, index_path: str, output_dir: str = "baseline"):
+        """Build FAISS flat baseline using the same embeddings as LEANN index"""
+        print("🔨 Building FAISS Flat baseline...")
+
+        import os
+        import pickle
+
+        import numpy as np
+        from leann.api import compute_embeddings
+        from leann_backend_hnsw import faiss
+
+        os.makedirs(output_dir, exist_ok=True)
+        baseline_path = os.path.join(output_dir, "faiss_flat.index")
+        metadata_path = os.path.join(output_dir, "metadata.pkl")
+
+        if os.path.exists(baseline_path) and os.path.exists(metadata_path):
+            print(f"✅ Baseline already exists at {baseline_path}")
+            return baseline_path
+
+        # Read metadata from the built index
+        meta_path = f"{index_path}.meta.json"
+        with open(meta_path) as f:
+            import json
+
+            meta = json.loads(f.read())
+
+        embedding_model = meta["embedding_model"]
+        passage_source = meta["passage_sources"][0]
+        passage_file = passage_source["path"]
+
+        # Convert relative path to absolute
+        if not os.path.isabs(passage_file):
+            index_dir = os.path.dirname(index_path)
+            passage_file = os.path.join(index_dir, os.path.basename(passage_file))
+
+        print(f"📊 Loading passages from {passage_file}...")
+        print(f"🤖 Using embedding model: {embedding_model}")
+
+        # Load all passages for baseline
+        passages = []
+        passage_ids = []
+        with open(passage_file, encoding="utf-8") as f:
+            for line in f:
+                if line.strip():
+                    data = json.loads(line)
+                    passages.append(data["text"])
+                    passage_ids.append(data["id"])
+
+        print(f"📄 Loaded {len(passages)} passages")
+
+        # Compute embeddings using the same method as LEANN
+        print("🧮 Computing embeddings...")
+        embeddings = compute_embeddings(
+            passages,
+            embedding_model,
+            mode="sentence-transformers",
+            use_server=False,
+        )
+
+        print(f"📐 Embedding shape: {embeddings.shape}")
+
+        # Build FAISS flat index
+        print("🏗️  Building FAISS IndexFlatIP...")
+        dimension = embeddings.shape[1]
+        index = faiss.IndexFlatIP(dimension)
+
+        # Add embeddings to flat index
+        embeddings_f32 = embeddings.astype(np.float32)
+        index.add(embeddings_f32.shape[0], faiss.swig_ptr(embeddings_f32))
+
+        # Save index and metadata
+        faiss.write_index(index, baseline_path)
+        with open(metadata_path, "wb") as f:
+            pickle.dump(passage_ids, f)
+
+        print(f"✅ FAISS baseline saved to {baseline_path}")
+        print(f"✅ Metadata saved to {metadata_path}")
+        print(f"📊 Total vectors: {index.ntotal}")
+
+        return baseline_path
+
+    def extract_pdf_text(self, pdf_path: Path) -> list[dict]:
+        """Extract and chunk text from a PDF file"""
+        chunks = []
+        doc = pymupdf.open(pdf_path)
+
+        for page_num in range(len(doc)):
+            page = doc[page_num]
+            text = page.get_text()  # type: ignore
+
+            if not text.strip():
+                continue
+
+            # Create metadata
+            metadata = {
+                "source_file": pdf_path.name,
+                "page_number": page_num + 1,
+                "document_type": "10K" if "10K" in pdf_path.name else "10Q",
+                "company": pdf_path.name.split("_")[0],
+                "doc_period": self.extract_year_from_filename(pdf_path.name),
+            }
+
+            # Use recursive character splitting like LangChain
+            if len(text.split()) > 500:
+                # Split by double newlines (paragraphs)
+                paragraphs = [p.strip() for p in text.split("\n\n") if p.strip()]
+
+                current_chunk = ""
+                for para in paragraphs:
+                    # If adding this paragraph would make chunk too long, save current chunk
+                    if current_chunk and len((current_chunk + " " + para).split()) > 300:
+                        if current_chunk.strip():
+                            chunks.append(
+                                {
+                                    "text": current_chunk.strip(),
+                                    "metadata": {
+                                        **metadata,
+                                        "chunk_id": f"page_{page_num + 1}_chunk_{len(chunks)}",
+                                    },
+                                }
+                            )
+                        current_chunk = para
+                    else:
+                        current_chunk = (current_chunk + " " + para).strip()
+
+                # Add the last chunk
+                if current_chunk.strip():
+                    chunks.append(
+                        {
+                            "text": current_chunk.strip(),
+                            "metadata": {
+                                **metadata,
+                                "chunk_id": f"page_{page_num + 1}_chunk_{len(chunks)}",
+                            },
+                        }
+                    )
+            else:
+                # Page is short enough, use as single chunk
+                chunks.append(
+                    {
+                        "text": text.strip(),
+                        "metadata": {**metadata, "chunk_id": f"page_{page_num + 1}"},
+                    }
+                )
+
+        doc.close()
+        return chunks
+
+    def extract_year_from_filename(self, filename: str) -> str:
+        """Extract year from PDF filename"""
+        # Try to find 4-digit year in filename
+
+        match = re.search(r"(\d{4})", filename)
+        return match.group(1) if match else "unknown"
+
+    def verify_setup(self, index_path: str):
+        """Verify the setup by testing a simple query"""
+        print("🧪 Verifying setup with test query...")
+
+        try:
+            searcher = LeannSearcher(index_path)
+
+            # Test query
+            test_query = "What is the capital expenditure for 3M in 2018?"
+            results = searcher.search(test_query, top_k=3)
+
+            print(f"✅ Test query successful! Found {len(results)} results:")
+            for i, result in enumerate(results, 1):
+                company = result.metadata.get("company", "Unknown")
+                year = result.metadata.get("doc_period", "Unknown")
+                page = result.metadata.get("page_number", "Unknown")
+                print(f"   {i}. {company} {year} (page {page}) - Score: {result.score:.3f}")
+                print(f"      {result.text[:100]}...")
+
+            searcher.cleanup()
+            print("✅ Setup verification completed successfully!")
+
+        except Exception as e:
+            print(f"❌ Setup verification failed: {e}")
+            raise
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Setup FinanceBench with full PDF datastore")
+    parser.add_argument("--data-dir", default="data", help="Data directory")
+    parser.add_argument(
+        "--backend", choices=["hnsw", "diskann"], default="hnsw", help="LEANN backend"
+    )
+    parser.add_argument(
+        "--embedding-model",
+        default="sentence-transformers/all-mpnet-base-v2",
+        help="Embedding model",
+    )
+    parser.add_argument("--max-workers", type=int, default=5, help="Parallel download workers")
+    parser.add_argument("--skip-download", action="store_true", help="Skip PDF download")
+    parser.add_argument("--skip-build", action="store_true", help="Skip index building")
+    parser.add_argument(
+        "--build-baseline-only",
+        action="store_true",
+        help="Only build FAISS baseline from existing index",
+    )
+
+    args = parser.parse_args()
+
+    print("🏦 FinanceBench Complete Setup")
+    print("=" * 50)
+
+    setup = FinanceBenchSetup(args.data_dir)
+
+    try:
+        if args.build_baseline_only:
+            # Only build baseline from existing index
+            index_path = setup.index_dir / f"financebench_full_{args.backend}"
+            index_file = f"{index_path}.index"
+            meta_file = f"{index_path}.leann.meta.json"
+
+            if not os.path.exists(index_file) or not os.path.exists(meta_file):
+                print("❌ Index files not found:")
+                print(f"   Index: {index_file}")
+                print(f"   Meta: {meta_file}")
+                print("💡 Run without --build-baseline-only to build the index first")
+                exit(1)
+
+            print(f"🔨 Building baseline from existing index: {index_path}")
+            baseline_path = setup.build_faiss_flat_baseline(str(index_path))
+            print(f"✅ Baseline built at {baseline_path}")
+            return
+
+        # Step 1: Download dataset
+        setup.download_dataset()
+
+        # Step 2: Download PDFs
+        if not args.skip_download:
+            setup.download_all_pdfs(max_workers=args.max_workers)
+        else:
+            print("⏭️  Skipping PDF download")
+
+        # Step 3: Build LEANN index
+        if not args.skip_build:
+            index_path = setup.build_leann_index(
+                backend=args.backend, embedding_model=args.embedding_model
+            )
+
+            # Step 4: Build FAISS flat baseline
+            print("\n🔨 Building FAISS flat baseline...")
+            baseline_path = setup.build_faiss_flat_baseline(index_path)
+            print(f"✅ Baseline built at {baseline_path}")
+
+            # Step 5: Verify setup
+            setup.verify_setup(index_path)
+        else:
+            print("⏭️  Skipping index building")
+
+        print("\n🎉 FinanceBench setup completed!")
+        print(f"📁 Data directory: {setup.data_dir.absolute()}")
+        print("\nNext steps:")
+        print(
+            "1. Run evaluation: python evaluate_financebench.py --index data/index/financebench_full_hnsw.leann"
+        )
+        print(
+            "2. Or test manually: python -c \"from leann import LeannSearcher; s = LeannSearcher('data/index/financebench_full_hnsw.leann'); print(s.search('3M capital expenditure 2018'))\""
+        )
+
+    except KeyboardInterrupt:
+        print("\n⚠️  Setup interrupted by user")
+        exit(1)
+    except Exception as e:
+        print(f"\n❌ Setup failed: {e}")
+        exit(1)
+
+
+if __name__ == "__main__":
+    main()

benchmarks/financebench/verify_recall.py

@@ -0,0 +1,214 @@
+#!/usr/bin/env python3
+# /// script
+# requires-python = ">=3.9"
+# dependencies = [
+#     "faiss-cpu",
+#     "numpy",
+#     "sentence-transformers",
+#     "torch",
+#     "tqdm",
+# ]
+# ///
+
+"""
+Independent recall verification script using standard FAISS.
+Creates two indexes (HNSW and Flat) and compares recall@3 at different complexities.
+"""
+
+import json
+import time
+from pathlib import Path
+
+import faiss
+import numpy as np
+from sentence_transformers import SentenceTransformer
+from tqdm import tqdm
+
+
+def compute_embeddings_direct(chunks: list[str], model_name: str) -> np.ndarray:
+    """
+    Direct embedding computation using sentence-transformers.
+    Copied logic to avoid dependency issues.
+    """
+    print(f"Loading model: {model_name}")
+    model = SentenceTransformer(model_name)
+
+    print(f"Computing embeddings for {len(chunks)} chunks...")
+    embeddings = model.encode(
+        chunks,
+        show_progress_bar=True,
+        batch_size=32,
+        convert_to_numpy=True,
+        normalize_embeddings=False,
+    )
+
+    return embeddings.astype(np.float32)
+
+
+def load_financebench_queries(dataset_path: str, max_queries: int = 200) -> list[str]:
+    """Load FinanceBench queries from dataset"""
+    queries = []
+    with open(dataset_path, encoding="utf-8") as f:
+        for line in f:
+            if line.strip():
+                data = json.loads(line)
+                queries.append(data["question"])
+                if len(queries) >= max_queries:
+                    break
+    return queries
+
+
+def load_passages_from_leann_index(index_path: str) -> tuple[list[str], list[str]]:
+    """Load passages from LEANN index structure"""
+    meta_path = f"{index_path}.meta.json"
+    with open(meta_path) as f:
+        meta = json.load(f)
+
+    passage_source = meta["passage_sources"][0]
+    passage_file = passage_source["path"]
+
+    # Convert relative path to absolute
+    if not Path(passage_file).is_absolute():
+        index_dir = Path(index_path).parent
+        passage_file = index_dir / Path(passage_file).name
+
+    print(f"Loading passages from {passage_file}")
+
+    passages = []
+    passage_ids = []
+    with open(passage_file, encoding="utf-8") as f:
+        for line in tqdm(f, desc="Loading passages"):
+            if line.strip():
+                data = json.loads(line)
+                passages.append(data["text"])
+                passage_ids.append(data["id"])
+
+    print(f"Loaded {len(passages)} passages")
+    return passages, passage_ids
+
+
+def build_faiss_indexes(embeddings: np.ndarray) -> tuple[faiss.Index, faiss.Index]:
+    """Build FAISS indexes: Flat (ground truth) and HNSW"""
+    dimension = embeddings.shape[1]
+
+    # Build Flat index (ground truth)
+    print("Building FAISS IndexFlatIP (ground truth)...")
+    flat_index = faiss.IndexFlatIP(dimension)
+    flat_index.add(embeddings)
+
+    # Build HNSW index
+    print("Building FAISS IndexHNSWFlat...")
+    M = 32  # Same as LEANN default
+    hnsw_index = faiss.IndexHNSWFlat(dimension, M, faiss.METRIC_INNER_PRODUCT)
+    hnsw_index.hnsw.efConstruction = 200  # Same as LEANN default
+    hnsw_index.add(embeddings)
+
+    print(f"Built indexes with {flat_index.ntotal} vectors, dimension {dimension}")
+    return flat_index, hnsw_index
+
+
+def evaluate_recall_at_k(
+    query_embeddings: np.ndarray,
+    flat_index: faiss.Index,
+    hnsw_index: faiss.Index,
+    passage_ids: list[str],
+    k: int = 3,
+    ef_search: int = 64,
+) -> float:
+    """Evaluate recall@k comparing HNSW vs Flat"""
+
+    # Set search parameters for HNSW
+    hnsw_index.hnsw.efSearch = ef_search
+
+    total_recall = 0.0
+    num_queries = query_embeddings.shape[0]
+
+    for i in range(num_queries):
+        query = query_embeddings[i : i + 1]  # Keep 2D shape
+
+        # Get ground truth from Flat index (standard FAISS API)
+        flat_distances, flat_indices = flat_index.search(query, k)
+        ground_truth_ids = {passage_ids[idx] for idx in flat_indices[0]}
+
+        # Get results from HNSW index (standard FAISS API)
+        hnsw_distances, hnsw_indices = hnsw_index.search(query, k)
+        hnsw_ids = {passage_ids[idx] for idx in hnsw_indices[0]}
+
+        # Calculate recall
+        intersection = ground_truth_ids.intersection(hnsw_ids)
+        recall = len(intersection) / k
+        total_recall += recall
+
+        if i < 3:  # Show first few examples
+            print(f"  Query {i + 1}: Recall@{k} = {recall:.3f}")
+            print(f"    Flat: {list(ground_truth_ids)}")
+            print(f"    HNSW: {list(hnsw_ids)}")
+            print(f"    Intersection: {list(intersection)}")
+
+    avg_recall = total_recall / num_queries
+    return avg_recall
+
+
+def main():
+    # Configuration
+    dataset_path = "data/financebench_merged.jsonl"
+    index_path = "data/index/financebench_full_hnsw.leann"
+    embedding_model = "sentence-transformers/all-mpnet-base-v2"
+
+    print("🔍 FAISS Recall Verification")
+    print("=" * 50)
+
+    # Check if files exist
+    if not Path(dataset_path).exists():
+        print(f"❌ Dataset not found: {dataset_path}")
+        return
+    if not Path(f"{index_path}.meta.json").exists():
+        print(f"❌ Index metadata not found: {index_path}.meta.json")
+        return
+
+    # Load data
+    print("📖 Loading FinanceBench queries...")
+    queries = load_financebench_queries(dataset_path, max_queries=50)
+    print(f"Loaded {len(queries)} queries")
+
+    print("📄 Loading passages from LEANN index...")
+    passages, passage_ids = load_passages_from_leann_index(index_path)
+
+    # Compute embeddings
+    print("🧮 Computing passage embeddings...")
+    passage_embeddings = compute_embeddings_direct(passages, embedding_model)
+
+    print("🧮 Computing query embeddings...")
+    query_embeddings = compute_embeddings_direct(queries, embedding_model)
+
+    # Build FAISS indexes
+    print("🏗️ Building FAISS indexes...")
+    flat_index, hnsw_index = build_faiss_indexes(passage_embeddings)
+
+    # Test different efSearch values (equivalent to LEANN complexity)
+    print("\n📊 Evaluating Recall@3 at different efSearch values...")
+    ef_search_values = [16, 32, 64, 128, 256]
+
+    for ef_search in ef_search_values:
+        print(f"\n🧪 Testing efSearch = {ef_search}")
+        start_time = time.time()
+
+        recall = evaluate_recall_at_k(
+            query_embeddings, flat_index, hnsw_index, passage_ids, k=3, ef_search=ef_search
+        )
+
+        elapsed = time.time() - start_time
+        print(
+            f"📈 efSearch {ef_search}: Recall@3 = {recall:.3f} ({recall * 100:.1f}%) in {elapsed:.2f}s"
+        )
+
+    print("\n✅ Verification completed!")
+    print("\n📋 Summary:")
+    print("  - Built independent FAISS Flat and HNSW indexes")
+    print("  - Compared recall@3 at different efSearch values")
+    print("  - Used same embedding model as LEANN")
+    print("  - This validates LEANN's recall measurements")
+
+
+if __name__ == "__main__":
+    main()

benchmarks/generation_speed_bench.py

@@ -1,114 +0,0 @@
-import argparse
-import re
-import sys
-import time
-from pathlib import Path
-from statistics import mean
-
-from leann.chat import get_llm
-
-
-def parse_prompts_from_file(file_path: str) -> list[str]:
-    """
-    Parse a prompt dump file into individual prompt strings.
-
-    Splits by lines that look like: "PROMPT #<n>:".
-    Keeps the content from each marker up to the next marker (or EOF).
-    """
-    with open(file_path, "r", encoding="utf-8") as f:
-        text = f.read()
-
-    matches = list(re.finditer(r"^PROMPT\s+#\d+:\s*$", text, flags=re.MULTILINE))
-    if not matches:
-        # Fallback: try a more permissive pattern
-        matches = list(
-            re.finditer(r"^=+\nPROMPT\s+#\d+:\n=+\s*$", text, flags=re.MULTILINE)
-        )
-
-    prompts: list[str] = []
-    if not matches:
-        # No explicit markers; treat the whole file as a single prompt
-        return [text]
-
-    for i, m in enumerate(matches):
-        start = m.end()
-        end = matches[i + 1].start() if i + 1 < len(matches) else len(text)
-        block = text[start:end].strip()
-        # Reattach the marker line content above the block for full context
-        header_line_start = text.rfind("\n", 0, m.start()) + 1
-        header = text[header_line_start : m.end()].strip()
-        prompts.append(f"{header}\n{block}".strip())
-
-    return prompts
-
-
-def main():
-    parser = argparse.ArgumentParser(
-        description=(
-            "Iterate prompts in a dump file, time generations, print outputs, and report last-10 average time."
-        )
-    )
-    parser.add_argument(
-        "--path",
-        default="benchmarks/data/prompts_g5/prompt_dump_nq_hnsw.txt",
-        help="Path to the prompt dump file",
-    )
-    parser.add_argument(
-        "--type",
-        default="ollama",
-        choices=["hf", "openai", "ollama", "gemini", "simulated"],
-        help="LLM backend type",
-    )
-    parser.add_argument(
-        "--model",
-        default="Qwen/Qwen3-4B",
-        help="Model identifier (depends on backend)",
-    )
-    parser.add_argument(
-        "--max_tokens",
-        type=int,
-        default=512,
-        help="Max new tokens to generate per prompt",
-    )
-    args = parser.parse_args()
-
-    llm_config = {"type": args.type, "model": args.model}
-    chat = get_llm(llm_config)
-
-    prompts = parse_prompts_from_file(args.path)
-    print(f"Found {len(prompts)} prompts in {args.path}")
-
-    times: list[float] = []
-    for idx, prompt in enumerate(prompts, start=1):
-        print("\n" + "=" * 80)
-        print(f"PROMPT {idx}/{len(prompts)}")
-        print("-" * 80)
-        start = time.perf_counter()
-        try:
-            output = chat.ask(prompt, max_tokens=args.max_tokens)
-        except Exception as e:
-            output = f"<error: {e}>"
-        elapsed = time.perf_counter() - start
-        times.append(elapsed)
-        print(f"Time: {elapsed:.3f}s")
-        print("-" * 80)
-        print(output)
-        print("=" * 80)
-
-    if times:
-        window = times[-10:] if len(times) >= 10 else times
-        avg_last_10 = mean(window)
-        print(
-            f"\nAverage time over last {len(window)} prompts: {avg_last_10:.3f}s"
-        )
-    else:
-        print("No prompts processed.")
-
-
-if __name__ == "__main__":
-    main()
-
-
-
-
-

benchmarks/laion/.gitignore

@@ -0,0 +1 @@
+data/

benchmarks/laion/README.md

@@ -0,0 +1,199 @@
+# LAION Multimodal Benchmark
+
+A multimodal benchmark for evaluating image retrieval and generation performance using LEANN with CLIP embeddings and Qwen2.5-VL for multimodal generation on LAION dataset subset.
+
+## Overview
+
+This benchmark evaluates:
+- **Image retrieval timing** using caption-based queries
+- **Recall@K performance** for image search
+- **Complexity analysis** across different search parameters
+- **Index size and storage efficiency**
+- **Multimodal generation** with Qwen2.5-VL for image understanding and description
+
+## Dataset Configuration
+
+- **Dataset**: LAION-400M subset (10,000 images)
+- **Embeddings**: Pre-computed CLIP ViT-B/32 (512 dimensions)
+- **Queries**: 200 random captions from the dataset
+- **Ground Truth**: Self-recall (query caption → original image)
+
+## Quick Start
+
+### 1. Setup the benchmark
+
+```bash
+cd benchmarks/laion
+python setup_laion.py --num-samples 10000 --num-queries 200
+```
+
+This will:
+- Create dummy LAION data (10K samples)
+- Generate CLIP embeddings (512-dim)
+- Build LEANN index with HNSW backend
+- Create 200 evaluation queries
+
+### 2. Run evaluation
+
+```bash
+# Run all evaluation stages
+python evaluate_laion.py --index data/laion_index.leann
+
+# Run specific stages
+python evaluate_laion.py --index data/laion_index.leann --stage 2  # Recall evaluation
+python evaluate_laion.py --index data/laion_index.leann --stage 3  # Complexity analysis
+python evaluate_laion.py --index data/laion_index.leann --stage 4  # Index comparison
+python evaluate_laion.py --index data/laion_index.leann --stage 5  # Multimodal generation
+
+# Multimodal generation with Qwen2.5-VL
+python evaluate_laion.py --index data/laion_index.leann --stage 5 --model-name Qwen/Qwen2.5-VL-7B-Instruct
+```
+
+### 3. Save results
+
+```bash
+python evaluate_laion.py --index data/laion_index.leann --output results.json
+```
+
+## Configuration Options
+
+### Setup Options
+```bash
+python setup_laion.py \
+  --num-samples 10000 \
+  --num-queries 200 \
+  --index-path data/laion_index.leann \
+  --backend hnsw
+```
+
+### Evaluation Options
+```bash
+python evaluate_laion.py \
+  --index data/laion_index.leann \
+  --queries data/evaluation_queries.jsonl \
+  --complexity 64 \
+  --top-k 3 \
+  --num-samples 100 \
+  --stage all
+```
+
+## Evaluation Stages
+
+### Stage 2: Recall Evaluation
+- Evaluates Recall@3 for multimodal retrieval
+- Compares LEANN vs FAISS baseline performance
+- Self-recall: query caption should retrieve original image
+
+### Stage 3: Complexity Analysis
+- Binary search for optimal complexity (90% recall target)
+- Tests performance across different complexity levels
+- Analyzes speed vs. accuracy tradeoffs
+
+### Stage 4: Index Comparison
+- Compares compact vs non-compact index sizes
+- Measures search performance differences
+- Reports storage efficiency and speed ratios
+
+### Stage 5: Multimodal Generation
+- Uses Qwen2.5-VL for image understanding and description
+- Retrieval-Augmented Generation (RAG) with multimodal context
+- Measures both search and generation timing
+
+## Output Metrics
+
+### Timing Metrics
+- Average/median/min/max search time
+- Standard deviation
+- Searches per second
+- Latency in milliseconds
+
+### Recall Metrics
+- Recall@3 percentage for image retrieval
+- Number of queries with ground truth
+
+### Index Metrics
+- Total index size (MB)
+- Component breakdown (index, passages, metadata)
+- Storage savings (compact vs non-compact)
+- Backend and embedding model info
+
+### Generation Metrics (Stage 5)
+- Average search time per query
+- Average generation time per query
+- Time distribution (search vs generation)
+- Sample multimodal responses
+- Model: Qwen2.5-VL performance
+
+## Benchmark Results
+
+### LEANN-RAG Performance (CLIP ViT-L/14 + Qwen2.5-VL)
+
+**Stage 3: Optimal Complexity Analysis**
+- **Optimal Complexity**: 85 (achieving 90% Recall@3)
+- **Binary Search Range**: 1-128
+- **Target Recall**: 90%
+- **Index Type**: Non-compact (for fast binary search)
+
+**Stage 5: Multimodal Generation Performance (Qwen2.5-VL)**
+- **Total Queries**: 20
+- **Average Search Time**: 1.200s per query
+- **Average Generation Time**: 6.558s per query
+- **Time Distribution**: Search 15.5%, Generation 84.5%
+- **LLM Backend**: HuggingFace transformers
+- **Model**: Qwen/Qwen2.5-VL-7B-Instruct
+- **Optimal Complexity**: 85
+
+**System Performance:**
+- **Index Size**: ~10,000 image embeddings from LAION subset
+- **Embedding Model**: CLIP ViT-L/14 (768 dimensions)
+- **Backend**: HNSW with cosine distance
+
+### Example Results
+
+```
+🎯 LAION MULTIMODAL BENCHMARK RESULTS
+============================================================
+
+📊 Multimodal Generation Results:
+  Total Queries: 20
+  Avg Search Time: 1.200s
+  Avg Generation Time: 6.558s
+  Time Distribution: Search 15.5%, Generation 84.5%
+  LLM Backend: HuggingFace transformers
+  Model: Qwen/Qwen2.5-VL-7B-Instruct
+
+⚙️ Optimal Complexity Analysis:
+  Target Recall: 90%
+  Optimal Complexity: 85
+  Binary Search Range: 1-128
+  Non-compact Index (fast search, no recompute)
+
+🚀 Performance Summary:
+  Multimodal RAG: 7.758s total per query
+  Search: 15.5% of total time
+  Generation: 84.5% of total time
+```
+
+## Directory Structure
+
+```
+benchmarks/laion/
+├── setup_laion.py           # Setup script
+├── evaluate_laion.py        # Evaluation script
+├── README.md               # This file
+└── data/                   # Generated data
+    ├── laion_images/       # Image files (placeholder)
+    ├── laion_metadata.jsonl # Image metadata
+    ├── laion_passages.jsonl # LEANN passages
+    ├── laion_embeddings.npy # CLIP embeddings
+    ├── evaluation_queries.jsonl # Evaluation queries
+    └── laion_index.leann/  # LEANN index files
+```
+
+## Notes
+
+- Current implementation uses dummy data for demonstration
+- For real LAION data, implement actual download logic in `setup_laion.py`
+- CLIP embeddings are randomly generated - replace with real CLIP model for production
+- Adjust `num_samples` and `num_queries` based on available resources
+- Consider using `--num-samples` during evaluation for faster testing

benchmarks/laion/evaluate_laion.py

@@ -0,0 +1,725 @@
+"""
+LAION Multimodal Benchmark Evaluation Script - Modular Recall-based Evaluation
+"""
+
+import argparse
+import json
+import logging
+import os
+import pickle
+import time
+from pathlib import Path
+
+import numpy as np
+from leann import LeannSearcher
+from leann_backend_hnsw import faiss
+from sentence_transformers import SentenceTransformer
+
+from ..llm_utils import evaluate_multimodal_rag, load_qwen_vl_model
+
+# Setup logging to reduce verbose output
+logging.basicConfig(level=logging.WARNING)
+logging.getLogger("leann.api").setLevel(logging.WARNING)
+logging.getLogger("leann_backend_hnsw").setLevel(logging.WARNING)
+
+
+class RecallEvaluator:
+    """Stage 2: Evaluate Recall@3 (LEANN vs FAISS baseline for multimodal retrieval)"""
+
+    def __init__(self, index_path: str, baseline_dir: str):
+        self.index_path = index_path
+        self.baseline_dir = baseline_dir
+        self.searcher = LeannSearcher(index_path)
+
+        # Load FAISS flat baseline (image embeddings)
+        baseline_index_path = os.path.join(baseline_dir, "faiss_flat.index")
+        metadata_path = os.path.join(baseline_dir, "metadata.pkl")
+
+        self.faiss_index = faiss.read_index(baseline_index_path)
+        with open(metadata_path, "rb") as f:
+            self.image_ids = pickle.load(f)
+        print(f"📚 Loaded FAISS flat baseline with {self.faiss_index.ntotal} image vectors")
+
+        # Load sentence-transformers CLIP for text embedding (ViT-L/14)
+        self.st_clip = SentenceTransformer("clip-ViT-L-14")
+
+    def evaluate_recall_at_3(
+        self, captions: list[str], complexity: int = 64, recompute_embeddings: bool = True
+    ) -> float:
+        """Evaluate recall@3 for multimodal retrieval: caption queries -> image results"""
+        recompute_str = "with recompute" if recompute_embeddings else "no recompute"
+        print(f"🔍 Evaluating recall@3 with complexity={complexity} ({recompute_str})...")
+
+        total_recall = 0.0
+        num_queries = len(captions)
+
+        for i, caption in enumerate(captions):
+            # Get ground truth: search with FAISS flat using caption text embedding
+            # Generate CLIP text embedding for caption via sentence-transformers (normalized)
+            query_embedding = self.st_clip.encode(
+                [caption], convert_to_numpy=True, normalize_embeddings=True, show_progress_bar=False
+            ).astype(np.float32)
+
+            # Search FAISS flat for ground truth using LEANN's modified faiss API
+            n = query_embedding.shape[0]  # Number of queries
+            k = 3  # Number of nearest neighbors
+            distances = np.zeros((n, k), dtype=np.float32)
+            labels = np.zeros((n, k), dtype=np.int64)
+
+            self.faiss_index.search(
+                n,
+                faiss.swig_ptr(query_embedding),
+                k,
+                faiss.swig_ptr(distances),
+                faiss.swig_ptr(labels),
+            )
+
+            # Extract the results (image IDs from FAISS)
+            baseline_ids = {self.image_ids[idx] for idx in labels[0]}
+
+            # Search with LEANN at specified complexity (using caption as text query)
+            test_results = self.searcher.search(
+                caption,
+                top_k=3,
+                complexity=complexity,
+                recompute_embeddings=recompute_embeddings,
+            )
+            test_ids = {result.id for result in test_results}
+
+            # Calculate recall@3 = |intersection| / |ground_truth|
+            intersection = test_ids.intersection(baseline_ids)
+            recall = len(intersection) / 3.0  # Ground truth size is 3
+            total_recall += recall
+
+            if i < 3:  # Show first few examples
+                print(f"  Query {i + 1}: '{caption[:50]}...' -> Recall@3: {recall:.3f}")
+                print(f"    FAISS ground truth: {list(baseline_ids)}")
+                print(f"    LEANN results (C={complexity}, {recompute_str}): {list(test_ids)}")
+                print(f"    Intersection: {list(intersection)}")
+
+        avg_recall = total_recall / num_queries
+        print(f"📊 Average Recall@3: {avg_recall:.3f} ({avg_recall * 100:.1f}%)")
+        return avg_recall
+
+    def cleanup(self):
+        """Cleanup resources"""
+        if hasattr(self, "searcher"):
+            self.searcher.cleanup()
+
+
+class LAIONEvaluator:
+    def __init__(self, index_path: str):
+        self.index_path = index_path
+        self.searcher = LeannSearcher(index_path)
+
+    def load_queries(self, queries_file: str) -> list[str]:
+        """Load caption queries from evaluation file"""
+        captions = []
+        with open(queries_file, encoding="utf-8") as f:
+            for line in f:
+                if line.strip():
+                    query_data = json.loads(line)
+                    captions.append(query_data["query"])
+
+        print(f"📊 Loaded {len(captions)} caption queries")
+        return captions
+
+    def analyze_index_sizes(self) -> dict:
+        """Analyze index sizes, emphasizing .index only (exclude passages)."""
+        print("📏 Analyzing index sizes (.index only)...")
+
+        # Get all index-related files
+        index_path = Path(self.index_path)
+        index_dir = index_path.parent
+        index_name = index_path.stem  # Remove .leann extension
+
+        sizes: dict[str, float] = {}
+
+        # Core index files
+        index_file = index_dir / f"{index_name}.index"
+        meta_file = index_dir / f"{index_path.name}.meta.json"  # Keep .leann for meta file
+        passages_file = index_dir / f"{index_path.name}.passages.jsonl"  # Keep .leann for passages
+        passages_idx_file = index_dir / f"{index_path.name}.passages.idx"  # Keep .leann for idx
+
+        # Core index size (.index only)
+        index_mb = index_file.stat().st_size / (1024 * 1024) if index_file.exists() else 0.0
+        sizes["index_only_mb"] = index_mb
+
+        # Other files for reference (not counted in index_only_mb)
+        sizes["metadata_mb"] = (
+            meta_file.stat().st_size / (1024 * 1024) if meta_file.exists() else 0.0
+        )
+        sizes["passages_text_mb"] = (
+            passages_file.stat().st_size / (1024 * 1024) if passages_file.exists() else 0.0
+        )
+        sizes["passages_index_mb"] = (
+            passages_idx_file.stat().st_size / (1024 * 1024) if passages_idx_file.exists() else 0.0
+        )
+
+        print(f"  📁 .index size: {index_mb:.1f} MB")
+        if sizes["metadata_mb"]:
+            print(f"  🧾 metadata: {sizes['metadata_mb']:.3f} MB")
+        if sizes["passages_text_mb"] or sizes["passages_index_mb"]:
+            print(
+                f"  (passages excluded) text: {sizes['passages_text_mb']:.1f} MB, idx: {sizes['passages_index_mb']:.1f} MB"
+            )
+
+        return sizes
+
+    def create_non_compact_index_for_comparison(self, non_compact_index_path: str) -> dict:
+        """Create a non-compact index for comparison purposes"""
+        print("🏗️ Building non-compact index from existing passages...")
+
+        # Load existing passages from current index
+        from leann import LeannBuilder
+
+        current_index_path = Path(self.index_path)
+        current_index_dir = current_index_path.parent
+        current_index_name = current_index_path.name
+
+        # Read metadata to get passage source
+        meta_path = current_index_dir / f"{current_index_name}.meta.json"
+        with open(meta_path) as f:
+            meta = json.load(f)
+
+        passage_source = meta["passage_sources"][0]
+        passage_file = passage_source["path"]
+
+        # Convert relative path to absolute
+        if not Path(passage_file).is_absolute():
+            passage_file = current_index_dir / Path(passage_file).name
+
+        print(f"📄 Loading passages from {passage_file}...")
+
+        # Load CLIP embeddings
+        embeddings_file = current_index_dir / "clip_image_embeddings.npy"
+        embeddings = np.load(embeddings_file)
+        print(f"📐 Loaded embeddings shape: {embeddings.shape}")
+
+        # Build non-compact index with same passages and embeddings
+        builder = LeannBuilder(
+            backend_name="hnsw",
+            # Use CLIP text encoder (ViT-L/14) to match image embeddings (768-dim)
+            embedding_model="clip-ViT-L-14",
+            embedding_mode="sentence-transformers",
+            is_recompute=False,  # Disable recompute (store embeddings)
+            is_compact=False,  # Disable compact storage
+            distance_metric="cosine",
+            **{
+                k: v
+                for k, v in meta.get("backend_kwargs", {}).items()
+                if k not in ["is_recompute", "is_compact", "distance_metric"]
+            },
+        )
+
+        # Prepare ids and add passages
+        ids: list[str] = []
+        with open(passage_file, encoding="utf-8") as f:
+            for line in f:
+                if line.strip():
+                    data = json.loads(line)
+                    ids.append(str(data["id"]))
+                    # Ensure metadata contains the id used by the vector index
+                    metadata = {**data.get("metadata", {}), "id": data["id"]}
+                    builder.add_text(text=data["text"], metadata=metadata)
+
+        if len(ids) != embeddings.shape[0]:
+            raise ValueError(
+                f"IDs count ({len(ids)}) does not match embeddings ({embeddings.shape[0]})."
+            )
+
+        # Persist a pickle for build_index_from_embeddings
+        pkl_path = current_index_dir / "clip_image_embeddings.pkl"
+        with open(pkl_path, "wb") as pf:
+            pickle.dump((ids, embeddings.astype(np.float32)), pf)
+
+        print(
+            f"🔨 Building non-compact index at {non_compact_index_path} from precomputed embeddings..."
+        )
+        builder.build_index_from_embeddings(non_compact_index_path, str(pkl_path))
+
+        # Analyze the non-compact index size
+        temp_evaluator = LAIONEvaluator(non_compact_index_path)
+        non_compact_sizes = temp_evaluator.analyze_index_sizes()
+        non_compact_sizes["index_type"] = "non_compact"
+
+        return non_compact_sizes
+
+    def compare_index_performance(
+        self, non_compact_path: str, compact_path: str, test_captions: list, complexity: int
+    ) -> dict:
+        """Compare performance between non-compact and compact indexes"""
+        print("⚡ Comparing search performance between indexes...")
+
+        # Test queries
+        test_queries = test_captions[:5]
+
+        results = {
+            "non_compact": {"search_times": []},
+            "compact": {"search_times": []},
+            "avg_search_times": {},
+            "speed_ratio": 0.0,
+        }
+
+        # Test non-compact index (no recompute)
+        print("  🔍 Testing non-compact index (no recompute)...")
+        non_compact_searcher = LeannSearcher(non_compact_path)
+
+        for caption in test_queries:
+            start_time = time.time()
+            _ = non_compact_searcher.search(
+                caption, top_k=3, complexity=complexity, recompute_embeddings=False
+            )
+            search_time = time.time() - start_time
+            results["non_compact"]["search_times"].append(search_time)
+
+        # Test compact index (with recompute)
+        print("  🔍 Testing compact index (with recompute)...")
+        compact_searcher = LeannSearcher(compact_path)
+
+        for caption in test_queries:
+            start_time = time.time()
+            _ = compact_searcher.search(
+                caption, top_k=3, complexity=complexity, recompute_embeddings=True
+            )
+            search_time = time.time() - start_time
+            results["compact"]["search_times"].append(search_time)
+
+        # Calculate averages
+        results["avg_search_times"]["non_compact"] = sum(
+            results["non_compact"]["search_times"]
+        ) / len(results["non_compact"]["search_times"])
+        results["avg_search_times"]["compact"] = sum(results["compact"]["search_times"]) / len(
+            results["compact"]["search_times"]
+        )
+
+        # Performance ratio
+        if results["avg_search_times"]["compact"] > 0:
+            results["speed_ratio"] = (
+                results["avg_search_times"]["non_compact"] / results["avg_search_times"]["compact"]
+            )
+        else:
+            results["speed_ratio"] = float("inf")
+
+        print(
+            f"    Non-compact (no recompute): {results['avg_search_times']['non_compact']:.3f}s avg"
+        )
+        print(f"    Compact (with recompute): {results['avg_search_times']['compact']:.3f}s avg")
+        print(f"    Speed ratio: {results['speed_ratio']:.2f}x")
+
+        # Cleanup
+        non_compact_searcher.cleanup()
+        compact_searcher.cleanup()
+
+        return results
+
+    def _print_results(self, timing_metrics: dict):
+        """Print evaluation results"""
+        print("\n🎯 LAION MULTIMODAL BENCHMARK RESULTS")
+        print("=" * 60)
+
+        # Index comparison analysis (prefer .index-only view if present)
+        if "current_index" in timing_metrics and "non_compact_index" in timing_metrics:
+            current = timing_metrics["current_index"]
+            non_compact = timing_metrics["non_compact_index"]
+
+            if "index_only_mb" in current and "index_only_mb" in non_compact:
+                print("\n📏 Index Comparison Analysis (.index only):")
+                print(f"  Compact index (current): {current.get('index_only_mb', 0):.1f} MB")
+                print(f"  Non-compact index: {non_compact.get('index_only_mb', 0):.1f} MB")
+                print(
+                    f"  Storage saving by compact: {timing_metrics.get('storage_saving_percent', 0):.1f}%"
+                )
+                # Show excluded components for reference if available
+                if any(
+                    k in non_compact
+                    for k in ("passages_text_mb", "passages_index_mb", "metadata_mb")
+                ):
+                    print("  (passages excluded in totals, shown for reference):")
+                    print(
+                        f"    - Passages text: {non_compact.get('passages_text_mb', 0):.1f} MB, "
+                        f"Passages index: {non_compact.get('passages_index_mb', 0):.1f} MB, "
+                        f"Metadata: {non_compact.get('metadata_mb', 0):.3f} MB"
+                    )
+            else:
+                # Fallback to legacy totals if running with older metrics
+                print("\n📏 Index Comparison Analysis:")
+                print(
+                    f"  Compact index (current): {current.get('total_with_embeddings', 0):.1f} MB"
+                )
+                print(
+                    f"  Non-compact index (with embeddings): {non_compact.get('total_with_embeddings', 0):.1f} MB"
+                )
+                print(
+                    f"  Storage saving by compact: {timing_metrics.get('storage_saving_percent', 0):.1f}%"
+                )
+                print("  Component breakdown (non-compact):")
+                print(f"    - Main index: {non_compact.get('index', 0):.1f} MB")
+                print(f"    - Passages text: {non_compact.get('passages_text', 0):.1f} MB")
+                print(f"    - Passages index: {non_compact.get('passages_index', 0):.1f} MB")
+                print(f"    - Metadata: {non_compact.get('metadata', 0):.1f} MB")
+
+        # Performance comparison
+        if "performance_comparison" in timing_metrics:
+            perf = timing_metrics["performance_comparison"]
+            print("\n⚡ Performance Comparison:")
+            print(
+                f"  Non-compact (no recompute): {perf.get('avg_search_times', {}).get('non_compact', 0):.3f}s avg"
+            )
+            print(
+                f"  Compact (with recompute): {perf.get('avg_search_times', {}).get('compact', 0):.3f}s avg"
+            )
+            print(f"  Speed ratio: {perf.get('speed_ratio', 0):.2f}x")
+
+        # Legacy single index analysis (fallback)
+        if "total_with_embeddings" in timing_metrics and "current_index" not in timing_metrics:
+            print("\n📏 Index Size Analysis:")
+            print(
+                f"  Index with embeddings: {timing_metrics.get('total_with_embeddings', 0):.1f} MB"
+            )
+            print(
+                f"  Estimated pruned index: {timing_metrics.get('total_without_embeddings', 0):.1f} MB"
+            )
+            print(f"  Compression ratio: {timing_metrics.get('compression_ratio', 0):.2f}x")
+
+    def cleanup(self):
+        """Cleanup resources"""
+        if self.searcher:
+            self.searcher.cleanup()
+
+
+def main():
+    parser = argparse.ArgumentParser(description="LAION Multimodal Benchmark Evaluation")
+    parser.add_argument("--index", required=True, help="Path to LEANN index")
+    parser.add_argument(
+        "--queries", default="data/evaluation_queries.jsonl", help="Path to evaluation queries"
+    )
+    parser.add_argument(
+        "--stage",
+        choices=["2", "3", "4", "5", "all"],
+        default="all",
+        help="Which stage to run (2=recall, 3=complexity, 4=index comparison, 5=generation)",
+    )
+    parser.add_argument("--complexity", type=int, default=None, help="Complexity for search")
+    parser.add_argument("--baseline-dir", default="baseline", help="Baseline output directory")
+    parser.add_argument("--output", help="Save results to JSON file")
+    parser.add_argument(
+        "--llm-backend",
+        choices=["hf"],
+        default="hf",
+        help="LLM backend (Qwen2.5-VL only supports HF)",
+    )
+    parser.add_argument(
+        "--model-name", default="Qwen/Qwen2.5-VL-7B-Instruct", help="Multimodal model name"
+    )
+
+    args = parser.parse_args()
+
+    try:
+        # Check if baseline exists
+        baseline_index_path = os.path.join(args.baseline_dir, "faiss_flat.index")
+        if not os.path.exists(baseline_index_path):
+            print(f"❌ FAISS baseline not found at {baseline_index_path}")
+            print("💡 Please run setup_laion.py first to build the baseline")
+            exit(1)
+
+        if args.stage == "2" or args.stage == "all":
+            # Stage 2: Recall@3 evaluation
+            print("🚀 Starting Stage 2: Recall@3 evaluation for multimodal retrieval")
+
+            evaluator = RecallEvaluator(args.index, args.baseline_dir)
+
+            # Load caption queries for testing
+            laion_evaluator = LAIONEvaluator(args.index)
+            captions = laion_evaluator.load_queries(args.queries)
+
+            # Test with queries for robust measurement
+            test_captions = captions[:100]  # Use subset for speed
+            print(f"🧪 Testing with {len(test_captions)} caption queries")
+
+            # Test with complexity 64
+            complexity = 64
+            recall = evaluator.evaluate_recall_at_3(test_captions, complexity)
+            print(f"📈 Recall@3 at complexity {complexity}: {recall * 100:.1f}%")
+
+            evaluator.cleanup()
+            print("✅ Stage 2 completed!\n")
+
+        # Shared non-compact index path for Stage 3 and 4
+        non_compact_index_path = args.index.replace(".leann", "_noncompact.leann")
+        complexity = args.complexity
+
+        if args.stage == "3" or args.stage == "all":
+            # Stage 3: Binary search for 90% recall complexity
+            print("🚀 Starting Stage 3: Binary search for 90% recall complexity")
+            print(
+                "💡 Creating non-compact index for fast binary search with recompute_embeddings=False"
+            )
+
+            # Create non-compact index for binary search
+            print("🏗️ Creating non-compact index for binary search...")
+            evaluator = LAIONEvaluator(args.index)
+            evaluator.create_non_compact_index_for_comparison(non_compact_index_path)
+
+            # Use non-compact index for binary search
+            binary_search_evaluator = RecallEvaluator(non_compact_index_path, args.baseline_dir)
+
+            # Load caption queries for testing
+            captions = evaluator.load_queries(args.queries)
+
+            # Use subset for robust measurement
+            test_captions = captions[:50]  # Smaller subset for binary search speed
+            print(f"🧪 Testing with {len(test_captions)} caption queries")
+
+            # Binary search for 90% recall complexity
+            target_recall = 0.9
+            min_complexity, max_complexity = 1, 128
+
+            print(f"🔍 Binary search for {target_recall * 100}% recall complexity...")
+            print(f"Search range: {min_complexity} to {max_complexity}")
+
+            best_complexity = None
+            best_recall = 0.0
+
+            while min_complexity <= max_complexity:
+                mid_complexity = (min_complexity + max_complexity) // 2
+
+                print(
+                    f"\n🧪 Testing complexity {mid_complexity} (no recompute, non-compact index)..."
+                )
+                # Use recompute_embeddings=False on non-compact index for fast binary search
+                recall = binary_search_evaluator.evaluate_recall_at_3(
+                    test_captions, mid_complexity, recompute_embeddings=False
+                )
+
+                print(
+                    f"  Complexity {mid_complexity}: Recall@3 = {recall:.3f} ({recall * 100:.1f}%)"
+                )
+
+                if recall >= target_recall:
+                    best_complexity = mid_complexity
+                    best_recall = recall
+                    max_complexity = mid_complexity - 1
+                    print("  ✅ Target reached! Searching for lower complexity...")
+                else:
+                    min_complexity = mid_complexity + 1
+                    print("  ❌ Below target. Searching for higher complexity...")
+
+            if best_complexity is not None:
+                print("\n🎯 Optimal complexity found!")
+                print(f"  Complexity: {best_complexity}")
+                print(f"  Recall@3: {best_recall:.3f} ({best_recall * 100:.1f}%)")
+
+                # Test a few complexities around the optimal one for verification
+                print("\n🔬 Verification test around optimal complexity:")
+                verification_complexities = [
+                    max(1, best_complexity - 2),
+                    max(1, best_complexity - 1),
+                    best_complexity,
+                    best_complexity + 1,
+                    best_complexity + 2,
+                ]
+
+                for complexity in verification_complexities:
+                    if complexity <= 512:  # reasonable upper bound
+                        recall = binary_search_evaluator.evaluate_recall_at_3(
+                            test_captions, complexity, recompute_embeddings=False
+                        )
+                        status = "✅" if recall >= target_recall else "❌"
+                        print(f"  {status} Complexity {complexity:3d}: {recall * 100:5.1f}%")
+
+                # Now test the optimal complexity with compact index and recompute for comparison
+                print(
+                    f"\n🔄 Testing optimal complexity {best_complexity} on compact index WITH recompute..."
+                )
+                compact_evaluator = RecallEvaluator(args.index, args.baseline_dir)
+                recall_with_recompute = compact_evaluator.evaluate_recall_at_3(
+                    test_captions[:10], best_complexity, recompute_embeddings=True
+                )
+                print(
+                    f"  ✅ Complexity {best_complexity} (compact index with recompute): {recall_with_recompute * 100:.1f}%"
+                )
+                complexity = best_complexity
+                print(
+                    f"  📊 Recall difference: {abs(best_recall - recall_with_recompute) * 100:.2f}%"
+                )
+                compact_evaluator.cleanup()
+            else:
+                print(f"\n❌ Could not find complexity achieving {target_recall * 100}% recall")
+                print("All tested complexities were below target.")
+
+            # Cleanup evaluators (keep non-compact index for Stage 4)
+            binary_search_evaluator.cleanup()
+            evaluator.cleanup()
+
+            print("✅ Stage 3 completed! Non-compact index saved for Stage 4.\n")
+
+        if args.stage == "4" or args.stage == "all":
+            # Stage 4: Index comparison (without LLM generation)
+            print("🚀 Starting Stage 4: Index comparison analysis")
+
+            # Use LAION evaluator for index comparison
+            evaluator = LAIONEvaluator(args.index)
+
+            # Load caption queries
+            captions = evaluator.load_queries(args.queries)
+
+            # Step 1: Analyze current (compact) index
+            print("\n📏 Analyzing current index (compact, pruned)...")
+            compact_size_metrics = evaluator.analyze_index_sizes()
+            compact_size_metrics["index_type"] = "compact"
+
+            # Step 2: Use existing non-compact index or create if needed
+            if Path(non_compact_index_path).exists():
+                print(
+                    f"\n📁 Using existing non-compact index from Stage 3: {non_compact_index_path}"
+                )
+                temp_evaluator = LAIONEvaluator(non_compact_index_path)
+                non_compact_size_metrics = temp_evaluator.analyze_index_sizes()
+                non_compact_size_metrics["index_type"] = "non_compact"
+            else:
+                print("\n🏗️ Creating non-compact index (with embeddings) for comparison...")
+                non_compact_size_metrics = evaluator.create_non_compact_index_for_comparison(
+                    non_compact_index_path
+                )
+
+            # Step 3: Compare index sizes (.index only)
+            print("\n📊 Index size comparison (.index only):")
+            print(
+                f"  Compact index (current): {compact_size_metrics.get('index_only_mb', 0):.1f} MB"
+            )
+            print(f"  Non-compact index: {non_compact_size_metrics.get('index_only_mb', 0):.1f} MB")
+
+            storage_saving = 0.0
+            if non_compact_size_metrics.get("index_only_mb", 0) > 0:
+                storage_saving = (
+                    (
+                        non_compact_size_metrics.get("index_only_mb", 0)
+                        - compact_size_metrics.get("index_only_mb", 0)
+                    )
+                    / non_compact_size_metrics.get("index_only_mb", 1)
+                    * 100
+                )
+            print(f"  Storage saving by compact: {storage_saving:.1f}%")
+
+            # Step 4: Performance comparison between the two indexes
+            if complexity is None:
+                raise ValueError("Complexity is required for index comparison")
+
+            print("\n⚡ Performance comparison between indexes...")
+            performance_metrics = evaluator.compare_index_performance(
+                non_compact_index_path, args.index, captions[:10], complexity=complexity
+            )
+
+            # Combine all metrics
+            combined_metrics = {
+                "current_index": compact_size_metrics,
+                "non_compact_index": non_compact_size_metrics,
+                "performance_comparison": performance_metrics,
+                "storage_saving_percent": storage_saving,
+            }
+
+            # Print comprehensive results
+            evaluator._print_results(combined_metrics)
+
+            # Save results if requested
+            if args.output:
+                print(f"\n💾 Saving results to {args.output}...")
+                with open(args.output, "w") as f:
+                    json.dump(combined_metrics, f, indent=2, default=str)
+                print(f"✅ Results saved to {args.output}")
+
+            evaluator.cleanup()
+            print("✅ Stage 4 completed!\n")
+
+        if args.stage in ("5", "all"):
+            print("🚀 Starting Stage 5: Multimodal generation with Qwen2.5-VL")
+            evaluator = LAIONEvaluator(args.index)
+            captions = evaluator.load_queries(args.queries)
+            test_captions = captions[: min(20, len(captions))]  # Use subset for generation
+
+            print(f"🧪 Testing multimodal generation with {len(test_captions)} queries")
+
+            # Load Qwen2.5-VL model
+            try:
+                print("Loading Qwen2.5-VL model...")
+                processor, model = load_qwen_vl_model(args.model_name)
+
+                # Run multimodal generation evaluation
+                complexity = args.complexity or 64
+                gen_results = evaluate_multimodal_rag(
+                    evaluator.searcher,
+                    test_captions,
+                    processor=processor,
+                    model=model,
+                    complexity=complexity,
+                )
+
+                print("\n📊 Multimodal Generation Results:")
+                print(f"  Total Queries: {len(test_captions)}")
+                print(f"  Avg Search Time: {gen_results['avg_search_time']:.3f}s")
+                print(f"  Avg Generation Time: {gen_results['avg_generation_time']:.3f}s")
+                total_time = gen_results["avg_search_time"] + gen_results["avg_generation_time"]
+                search_pct = (gen_results["avg_search_time"] / total_time) * 100
+                gen_pct = (gen_results["avg_generation_time"] / total_time) * 100
+                print(f"  Time Distribution: Search {search_pct:.1f}%, Generation {gen_pct:.1f}%")
+                print("  LLM Backend: HuggingFace transformers")
+                print(f"  Model: {args.model_name}")
+
+                # Show sample results
+                print("\n📝 Sample Multimodal Generations:")
+                for i, response in enumerate(gen_results["results"][:3]):
+                    # Handle both string and dict formats for captions
+                    if isinstance(test_captions[i], dict):
+                        caption_text = test_captions[i].get("query", str(test_captions[i]))
+                    else:
+                        caption_text = str(test_captions[i])
+                    print(f"  Query {i + 1}: {caption_text[:60]}...")
+                    print(f"  Response {i + 1}: {response[:100]}...")
+                    print()
+
+            except Exception as e:
+                print(f"❌ Multimodal generation evaluation failed: {e}")
+                print("💡 Make sure transformers and Qwen2.5-VL are installed")
+                import traceback
+
+                traceback.print_exc()
+
+            evaluator.cleanup()
+            print("✅ Stage 5 completed!\n")
+
+        if args.stage == "all":
+            print("🎉 All evaluation stages completed successfully!")
+            print("\n📋 Summary:")
+            print("  Stage 2: ✅ Multimodal Recall@3 evaluation completed")
+            print("  Stage 3: ✅ Optimal complexity found")
+            print("  Stage 4: ✅ Index comparison analysis completed")
+            print("  Stage 5: ✅ Multimodal generation evaluation completed")
+            print("\n🔧 Recommended next steps:")
+            print("  - Use optimal complexity for best speed/accuracy balance")
+            print("  - Review index comparison for storage vs performance tradeoffs")
+
+            # Clean up non-compact index after all stages complete
+            print("\n🧹 Cleaning up temporary non-compact index...")
+            if Path(non_compact_index_path).exists():
+                temp_index_dir = Path(non_compact_index_path).parent
+                temp_index_name = Path(non_compact_index_path).name
+                for temp_file in temp_index_dir.glob(f"{temp_index_name}*"):
+                    temp_file.unlink()
+                print(f"✅ Cleaned up {non_compact_index_path}")
+            else:
+                print("📝 No temporary index to clean up")
+
+    except KeyboardInterrupt:
+        print("\n⚠️  Evaluation interrupted by user")
+        exit(1)
+    except Exception as e:
+        print(f"\n❌ Stage {args.stage} failed: {e}")
+        import traceback
+
+        traceback.print_exc()
+        exit(1)
+
+
+if __name__ == "__main__":
+    main()

benchmarks/laion/setup_laion.py

@@ -0,0 +1,576 @@
+"""
+LAION Multimodal Benchmark Setup Script
+Downloads LAION subset and builds LEANN index with sentence embeddings
+"""
+
+import argparse
+import asyncio
+import io
+import json
+import os
+import pickle
+import time
+from pathlib import Path
+
+import aiohttp
+import numpy as np
+from datasets import load_dataset
+from leann import LeannBuilder
+from PIL import Image
+from sentence_transformers import SentenceTransformer
+from tqdm import tqdm
+
+
+class LAIONSetup:
+    def __init__(self, data_dir: str = "data"):
+        self.data_dir = Path(data_dir)
+        self.images_dir = self.data_dir / "laion_images"
+        self.metadata_file = self.data_dir / "laion_metadata.jsonl"
+
+        # Create directories
+        self.data_dir.mkdir(exist_ok=True)
+        self.images_dir.mkdir(exist_ok=True)
+
+    async def download_single_image(self, session, sample_data, semaphore, progress_bar):
+        """Download a single image asynchronously"""
+        async with semaphore:  # Limit concurrent downloads
+            try:
+                image_url = sample_data["url"]
+                image_path = sample_data["image_path"]
+
+                # Skip if already exists
+                if os.path.exists(image_path):
+                    progress_bar.update(1)
+                    return sample_data
+
+                async with session.get(image_url, timeout=10) as response:
+                    if response.status == 200:
+                        content = await response.read()
+
+                        # Verify it's a valid image
+                        try:
+                            img = Image.open(io.BytesIO(content))
+                            img = img.convert("RGB")
+                            img.save(image_path, "JPEG")
+                            progress_bar.update(1)
+                            return sample_data
+                        except Exception:
+                            progress_bar.update(1)
+                            return None  # Skip invalid images
+                    else:
+                        progress_bar.update(1)
+                        return None
+
+            except Exception:
+                progress_bar.update(1)
+                return None
+
+    def download_laion_subset(self, num_samples: int = 1000):
+        """Download LAION subset from HuggingFace datasets with async parallel downloading"""
+        print(f"📥 Downloading LAION subset ({num_samples} samples)...")
+
+        # Load LAION-400M subset from HuggingFace
+        print("🤗 Loading from HuggingFace datasets...")
+        dataset = load_dataset("laion/laion400m", split="train", streaming=True)
+
+        # Collect sample metadata first (fast)
+        print("📋 Collecting sample metadata...")
+        candidates = []
+        for sample in dataset:
+            if len(candidates) >= num_samples * 3:  # Get 3x more candidates in case some fail
+                break
+
+            image_url = sample.get("url", "")
+            caption = sample.get("caption", "")
+
+            if not image_url or not caption:
+                continue
+
+            image_filename = f"laion_{len(candidates):06d}.jpg"
+            image_path = self.images_dir / image_filename
+
+            candidate = {
+                "id": f"laion_{len(candidates):06d}",
+                "url": image_url,
+                "caption": caption,
+                "image_path": str(image_path),
+                "width": sample.get("original_width", 512),
+                "height": sample.get("original_height", 512),
+                "similarity": sample.get("similarity", 0.0),
+            }
+            candidates.append(candidate)
+
+        print(
+            f"📊 Collected {len(candidates)} candidates, downloading {num_samples} in parallel..."
+        )
+
+        # Download images in parallel
+        async def download_batch():
+            semaphore = asyncio.Semaphore(20)  # Limit to 20 concurrent downloads
+            connector = aiohttp.TCPConnector(limit=100, limit_per_host=20)
+            timeout = aiohttp.ClientTimeout(total=30)
+
+            progress_bar = tqdm(total=len(candidates[: num_samples * 2]), desc="Downloading images")
+
+            async with aiohttp.ClientSession(connector=connector, timeout=timeout) as session:
+                tasks = []
+                for candidate in candidates[: num_samples * 2]:  # Try 2x more than needed
+                    task = self.download_single_image(session, candidate, semaphore, progress_bar)
+                    tasks.append(task)
+
+                # Wait for all downloads
+                results = await asyncio.gather(*tasks, return_exceptions=True)
+                progress_bar.close()
+
+                # Filter successful downloads
+                successful = [r for r in results if r is not None and not isinstance(r, Exception)]
+                return successful[:num_samples]
+
+        # Run async download
+        loop = asyncio.new_event_loop()
+        asyncio.set_event_loop(loop)
+        try:
+            samples = loop.run_until_complete(download_batch())
+        finally:
+            loop.close()
+
+        # Save metadata
+        with open(self.metadata_file, "w", encoding="utf-8") as f:
+            for sample in samples:
+                f.write(json.dumps(sample) + "\n")
+
+        print(f"✅ Downloaded {len(samples)} real LAION samples with async parallel downloading")
+        return samples
+
+    def generate_clip_image_embeddings(self, samples: list[dict]):
+        """Generate CLIP image embeddings for downloaded images"""
+        print("🔍 Generating CLIP image embeddings...")
+
+        # Load sentence-transformers CLIP (ViT-L/14, 768-dim) for image embeddings
+        # This single model can encode both images and text.
+        model = SentenceTransformer("clip-ViT-L-14")
+
+        embeddings = []
+        valid_samples = []
+
+        for sample in tqdm(samples, desc="Processing images"):
+            try:
+                # Load image
+                image_path = sample["image_path"]
+                image = Image.open(image_path).convert("RGB")
+
+                # Encode image to 768-dim embedding via sentence-transformers (normalized)
+                vec = model.encode(
+                    [image],
+                    convert_to_numpy=True,
+                    normalize_embeddings=True,
+                    batch_size=1,
+                    show_progress_bar=False,
+                )[0]
+                embeddings.append(vec.astype(np.float32))
+                valid_samples.append(sample)
+
+            except Exception as e:
+                print(f"  ⚠️ Failed to process {sample['id']}: {e}")
+                # Skip invalid images
+
+        embeddings = np.array(embeddings, dtype=np.float32)
+
+        # Save embeddings
+        embeddings_file = self.data_dir / "clip_image_embeddings.npy"
+        np.save(embeddings_file, embeddings)
+        print(f"✅ Generated {len(embeddings)} image embeddings, shape: {embeddings.shape}")
+
+        return embeddings, valid_samples
+
+    def build_faiss_baseline(
+        self, embeddings: np.ndarray, samples: list[dict], output_dir: str = "baseline"
+    ):
+        """Build FAISS flat baseline using CLIP image embeddings"""
+        print("🔨 Building FAISS Flat baseline...")
+
+        from leann_backend_hnsw import faiss
+
+        os.makedirs(output_dir, exist_ok=True)
+        baseline_path = os.path.join(output_dir, "faiss_flat.index")
+        metadata_path = os.path.join(output_dir, "metadata.pkl")
+
+        if os.path.exists(baseline_path) and os.path.exists(metadata_path):
+            print(f"✅ Baseline already exists at {baseline_path}")
+            return baseline_path
+
+        # Extract image IDs (must be present)
+        if not samples or "id" not in samples[0]:
+            raise KeyError("samples missing 'id' field for FAISS baseline")
+        image_ids: list[str] = [str(sample["id"]) for sample in samples]
+
+        print(f"📐 Embedding shape: {embeddings.shape}")
+        print(f"📄 Processing {len(image_ids)} images")
+
+        # Build FAISS flat index
+        print("🏗️ Building FAISS IndexFlatIP...")
+        dimension = embeddings.shape[1]
+        index = faiss.IndexFlatIP(dimension)
+
+        # Add embeddings to flat index
+        embeddings_f32 = embeddings.astype(np.float32)
+        index.add(embeddings_f32.shape[0], faiss.swig_ptr(embeddings_f32))
+
+        # Save index and metadata
+        faiss.write_index(index, baseline_path)
+        with open(metadata_path, "wb") as f:
+            pickle.dump(image_ids, f)
+
+        print(f"✅ FAISS baseline saved to {baseline_path}")
+        print(f"✅ Metadata saved to {metadata_path}")
+        print(f"📊 Total vectors: {index.ntotal}")
+
+        return baseline_path
+
+    def create_leann_passages(self, samples: list[dict]):
+        """Create LEANN-compatible passages from LAION data"""
+        print("📝 Creating LEANN passages...")
+
+        passages_file = self.data_dir / "laion_passages.jsonl"
+
+        with open(passages_file, "w", encoding="utf-8") as f:
+            for i, sample in enumerate(samples):
+                passage = {
+                    "id": sample["id"],
+                    "text": sample["caption"],  # Use caption as searchable text
+                    "metadata": {
+                        "image_url": sample["url"],
+                        "image_path": sample.get("image_path", ""),
+                        "width": sample["width"],
+                        "height": sample["height"],
+                        "similarity": sample["similarity"],
+                        "image_index": i,  # Index for embedding lookup
+                    },
+                }
+                f.write(json.dumps(passage) + "\n")
+
+        print(f"✅ Created {len(samples)} passages")
+        return passages_file
+
+    def build_compact_index(
+        self, passages_file: Path, embeddings: np.ndarray, index_path: str, backend: str = "hnsw"
+    ):
+        """Build compact LEANN index with CLIP embeddings (recompute=True, compact=True)"""
+        print(f"🏗️ Building compact LEANN index with {backend} backend...")
+
+        start_time = time.time()
+
+        # Save CLIP embeddings (npy) and also a pickle with (ids, embeddings)
+        npy_path = self.data_dir / "clip_image_embeddings.npy"
+        np.save(npy_path, embeddings)
+        print(f"💾 Saved CLIP embeddings to {npy_path}")
+
+        # Prepare ids in the same order as passages_file (matches embeddings order)
+        ids: list[str] = []
+        with open(passages_file, encoding="utf-8") as f:
+            for line in f:
+                if line.strip():
+                    rec = json.loads(line)
+                    ids.append(str(rec["id"]))
+
+        if len(ids) != embeddings.shape[0]:
+            raise ValueError(
+                f"IDs count ({len(ids)}) does not match embeddings ({embeddings.shape[0]})."
+            )
+
+        pkl_path = self.data_dir / "clip_image_embeddings.pkl"
+        with open(pkl_path, "wb") as pf:
+            pickle.dump((ids, embeddings.astype(np.float32)), pf)
+        print(f"💾 Saved (ids, embeddings) pickle to {pkl_path}")
+
+        # Initialize builder - compact with recompute
+        # Note: For multimodal case, we need to handle embeddings differently
+        # Let's try using sentence-transformers mode but with custom embeddings
+        builder = LeannBuilder(
+            backend_name=backend,
+            # Use CLIP text encoder (ViT-L/14) to match image space (768-dim)
+            embedding_model="clip-ViT-L-14",
+            embedding_mode="sentence-transformers",
+            # HNSW params (or forwarded to chosen backend)
+            graph_degree=32,
+            complexity=64,
+            # Compact/pruned with recompute at query time
+            is_recompute=True,
+            is_compact=True,
+            distance_metric="cosine",  # CLIP uses normalized vectors; cosine is appropriate
+            num_threads=4,
+        )
+
+        # Add passages (text + metadata)
+        print("📚 Adding passages...")
+        self._add_passages_with_embeddings(builder, passages_file, embeddings)
+
+        print(f"🔨 Building compact index at {index_path} from precomputed embeddings...")
+        builder.build_index_from_embeddings(index_path, str(pkl_path))
+
+        build_time = time.time() - start_time
+        print(f"✅ Compact index built in {build_time:.2f}s")
+
+        # Analyze index size
+        self._analyze_index_size(index_path)
+
+        return index_path
+
+    def build_non_compact_index(
+        self, passages_file: Path, embeddings: np.ndarray, index_path: str, backend: str = "hnsw"
+    ):
+        """Build non-compact LEANN index with CLIP embeddings (recompute=False, compact=False)"""
+        print(f"🏗️ Building non-compact LEANN index with {backend} backend...")
+
+        start_time = time.time()
+
+        # Ensure embeddings are saved (npy + pickle)
+        npy_path = self.data_dir / "clip_image_embeddings.npy"
+        if not npy_path.exists():
+            np.save(npy_path, embeddings)
+            print(f"💾 Saved CLIP embeddings to {npy_path}")
+        # Prepare ids in same order as passages_file
+        ids: list[str] = []
+        with open(passages_file, encoding="utf-8") as f:
+            for line in f:
+                if line.strip():
+                    rec = json.loads(line)
+                    ids.append(str(rec["id"]))
+        if len(ids) != embeddings.shape[0]:
+            raise ValueError(
+                f"IDs count ({len(ids)}) does not match embeddings ({embeddings.shape[0]})."
+            )
+        pkl_path = self.data_dir / "clip_image_embeddings.pkl"
+        if not pkl_path.exists():
+            with open(pkl_path, "wb") as pf:
+                pickle.dump((ids, embeddings.astype(np.float32)), pf)
+            print(f"💾 Saved (ids, embeddings) pickle to {pkl_path}")
+
+        # Initialize builder - non-compact without recompute
+        builder = LeannBuilder(
+            backend_name=backend,
+            embedding_model="clip-ViT-L-14",
+            embedding_mode="sentence-transformers",
+            graph_degree=32,
+            complexity=64,
+            is_recompute=False,  # Store embeddings (no recompute needed)
+            is_compact=False,  # Store full index (not pruned)
+            distance_metric="cosine",
+            num_threads=4,
+        )
+
+        # Add passages - embeddings will be loaded from file
+        print("📚 Adding passages...")
+        self._add_passages_with_embeddings(builder, passages_file, embeddings)
+
+        print(f"🔨 Building non-compact index at {index_path} from precomputed embeddings...")
+        builder.build_index_from_embeddings(index_path, str(pkl_path))
+
+        build_time = time.time() - start_time
+        print(f"✅ Non-compact index built in {build_time:.2f}s")
+
+        # Analyze index size
+        self._analyze_index_size(index_path)
+
+        return index_path
+
+    def _add_passages_with_embeddings(self, builder, passages_file: Path, embeddings: np.ndarray):
+        """Helper to add passages with pre-computed CLIP embeddings"""
+        with open(passages_file, encoding="utf-8") as f:
+            for line in tqdm(f, desc="Adding passages"):
+                if line.strip():
+                    passage = json.loads(line)
+
+                    # Add image metadata - LEANN will handle embeddings separately
+                    # Note: We store image metadata and caption text for searchability
+                    # Important: ensure passage ID in metadata matches vector ID
+                    builder.add_text(
+                        text=passage["text"],  # Image caption for searchability
+                        metadata={**passage["metadata"], "id": passage["id"]},
+                    )
+
+    def _analyze_index_size(self, index_path: str):
+        """Analyze index file sizes"""
+        print("📏 Analyzing index sizes...")
+
+        index_path = Path(index_path)
+        index_dir = index_path.parent
+        index_name = index_path.name  # e.g., laion_index.leann
+        index_prefix = index_path.stem  # e.g., laion_index
+
+        files = [
+            (f"{index_prefix}.index", ".index", "core"),
+            (f"{index_name}.meta.json", ".meta.json", "core"),
+            (f"{index_name}.ids.txt", ".ids.txt", "core"),
+            (f"{index_name}.passages.jsonl", ".passages.jsonl", "passages"),
+            (f"{index_name}.passages.idx", ".passages.idx", "passages"),
+        ]
+
+        def _fmt_size(bytes_val: int) -> str:
+            if bytes_val < 1024:
+                return f"{bytes_val} B"
+            kb = bytes_val / 1024
+            if kb < 1024:
+                return f"{kb:.1f} KB"
+            mb = kb / 1024
+            if mb < 1024:
+                return f"{mb:.2f} MB"
+            gb = mb / 1024
+            return f"{gb:.2f} GB"
+
+        total_index_only_mb = 0.0
+        total_all_mb = 0.0
+        for filename, label, group in files:
+            file_path = index_dir / filename
+            if file_path.exists():
+                size_bytes = file_path.stat().st_size
+                print(f"  {label}: {_fmt_size(size_bytes)}")
+                size_mb = size_bytes / (1024 * 1024)
+                total_all_mb += size_mb
+                if group == "core":
+                    total_index_only_mb += size_mb
+            else:
+                print(f"  {label}: (missing)")
+        print(f"  Total (index only, exclude passages): {total_index_only_mb:.2f} MB")
+        print(f"  Total (including passages): {total_all_mb:.2f} MB")
+
+    def create_evaluation_queries(self, samples: list[dict], num_queries: int = 200):
+        """Create evaluation queries from captions"""
+        print(f"📝 Creating {num_queries} evaluation queries...")
+
+        # Sample random captions as queries
+        import random
+
+        random.seed(42)  # For reproducibility
+
+        query_samples = random.sample(samples, min(num_queries, len(samples)))
+
+        queries_file = self.data_dir / "evaluation_queries.jsonl"
+        with open(queries_file, "w", encoding="utf-8") as f:
+            for sample in query_samples:
+                query = {
+                    "id": sample["id"],
+                    "query": sample["caption"],
+                    "ground_truth_id": sample["id"],  # For potential recall evaluation
+                }
+                f.write(json.dumps(query) + "\n")
+
+        print(f"✅ Created {len(query_samples)} evaluation queries")
+        return queries_file
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Setup LAION Multimodal Benchmark")
+    parser.add_argument("--data-dir", default="data", help="Data directory")
+    parser.add_argument("--num-samples", type=int, default=1000, help="Number of LAION samples")
+    parser.add_argument("--num-queries", type=int, default=50, help="Number of evaluation queries")
+    parser.add_argument("--index-path", default="data/laion_index.leann", help="Output index path")
+    parser.add_argument(
+        "--backend", default="hnsw", choices=["hnsw", "diskann"], help="LEANN backend"
+    )
+    parser.add_argument("--skip-download", action="store_true", help="Skip LAION dataset download")
+    parser.add_argument("--skip-build", action="store_true", help="Skip index building")
+
+    args = parser.parse_args()
+
+    print("🚀 Setting up LAION Multimodal Benchmark")
+    print("=" * 50)
+
+    try:
+        # Initialize setup
+        setup = LAIONSetup(args.data_dir)
+
+        # Step 1: Download LAION subset
+        if not args.skip_download:
+            print("\n📦 Step 1: Download LAION subset")
+            samples = setup.download_laion_subset(args.num_samples)
+
+            # Step 2: Generate CLIP image embeddings
+            print("\n🔍 Step 2: Generate CLIP image embeddings")
+            embeddings, valid_samples = setup.generate_clip_image_embeddings(samples)
+
+            # Step 3: Create LEANN passages (image metadata with embeddings)
+            print("\n📝 Step 3: Create LEANN passages")
+            passages_file = setup.create_leann_passages(valid_samples)
+        else:
+            print("⏭️  Skipping LAION dataset download")
+            # Load existing data
+            passages_file = setup.data_dir / "laion_passages.jsonl"
+            embeddings_file = setup.data_dir / "clip_image_embeddings.npy"
+
+            if not passages_file.exists() or not embeddings_file.exists():
+                raise FileNotFoundError(
+                    "Passages or embeddings file not found. Run without --skip-download first."
+                )
+
+            embeddings = np.load(embeddings_file)
+            print(f"📊 Loaded {len(embeddings)} embeddings from {embeddings_file}")
+
+        # Step 4: Build LEANN indexes (both compact and non-compact)
+        if not args.skip_build:
+            print("\n🏗️ Step 4: Build LEANN indexes with CLIP image embeddings")
+
+            # Build compact index (production mode - small, recompute required)
+            compact_index_path = args.index_path
+            print(f"Building compact index: {compact_index_path}")
+            setup.build_compact_index(passages_file, embeddings, compact_index_path, args.backend)
+
+            # Build non-compact index (comparison mode - large, fast search)
+            non_compact_index_path = args.index_path.replace(".leann", "_noncompact.leann")
+            print(f"Building non-compact index: {non_compact_index_path}")
+            setup.build_non_compact_index(
+                passages_file, embeddings, non_compact_index_path, args.backend
+            )
+
+            # Step 5: Build FAISS flat baseline
+            print("\n🔨 Step 5: Build FAISS flat baseline")
+            if not args.skip_download:
+                baseline_path = setup.build_faiss_baseline(embeddings, valid_samples)
+            else:
+                # Load valid_samples from passages file for FAISS baseline
+                valid_samples = []
+                with open(passages_file, encoding="utf-8") as f:
+                    for line in f:
+                        if line.strip():
+                            passage = json.loads(line)
+                            valid_samples.append({"id": passage["id"], "caption": passage["text"]})
+                baseline_path = setup.build_faiss_baseline(embeddings, valid_samples)
+
+            # Step 6: Create evaluation queries
+            print("\n📝 Step 6: Create evaluation queries")
+            queries_file = setup.create_evaluation_queries(valid_samples, args.num_queries)
+        else:
+            print("⏭️  Skipping index building")
+            baseline_path = "data/baseline/faiss_index.bin"
+            queries_file = setup.data_dir / "evaluation_queries.jsonl"
+
+        print("\n🎉 Setup completed successfully!")
+        print("📊 Summary:")
+        if not args.skip_download:
+            print(f"  Downloaded samples: {len(samples)}")
+            print(f"  Valid samples with embeddings: {len(valid_samples)}")
+        else:
+            print(f"  Loaded {len(embeddings)} embeddings")
+
+        if not args.skip_build:
+            print(f"  Compact index: {compact_index_path}")
+            print(f"  Non-compact index: {non_compact_index_path}")
+            print(f"  FAISS baseline: {baseline_path}")
+            print(f"  Queries: {queries_file}")
+
+            print("\n🔧 Next steps:")
+            print(f"  Run evaluation: python evaluate_laion.py --index {compact_index_path}")
+            print(f"  Or compare with: python evaluate_laion.py --index {non_compact_index_path}")
+        else:
+            print("  Skipped building indexes")
+
+    except KeyboardInterrupt:
+        print("\n⚠️  Setup interrupted by user")
+        exit(1)
+    except Exception as e:
+        print(f"\n❌ Setup failed: {e}")
+        exit(1)
+
+
+if __name__ == "__main__":
+    main()

benchmarks/llm_utils.py

@@ -0,0 +1,342 @@
+"""
+LLM utils for RAG benchmarks with Qwen3-8B and Qwen2.5-VL (multimodal)
+"""
+
+import time
+
+try:
+    import torch
+    from transformers import AutoModelForCausalLM, AutoTokenizer
+
+    HF_AVAILABLE = True
+except ImportError:
+    HF_AVAILABLE = False
+
+try:
+    from vllm import LLM, SamplingParams
+
+    VLLM_AVAILABLE = True
+except ImportError:
+    VLLM_AVAILABLE = False
+
+
+def is_qwen3_model(model_name):
+    """Check if model is Qwen3"""
+    return "Qwen3" in model_name or "qwen3" in model_name.lower()
+
+
+def is_qwen_vl_model(model_name):
+    """Check if model is Qwen2.5-VL"""
+    return "Qwen2.5-VL" in model_name or "qwen2.5-vl" in model_name.lower()
+
+
+def apply_qwen3_chat_template(tokenizer, prompt):
+    """Apply Qwen3 chat template with thinking enabled"""
+    messages = [{"role": "user", "content": prompt}]
+    return tokenizer.apply_chat_template(
+        messages,
+        tokenize=False,
+        add_generation_prompt=True,
+        enable_thinking=True,
+    )
+
+
+def extract_thinking_answer(response):
+    """Extract final answer from Qwen3 thinking model response"""
+    if "<think>" in response and "</think>" in response:
+        try:
+            think_end = response.index("</think>") + len("</think>")
+            final_answer = response[think_end:].strip()
+            return final_answer
+        except (ValueError, IndexError):
+            pass
+
+    return response.strip()
+
+
+def load_hf_model(model_name="Qwen/Qwen3-8B", trust_remote_code=False):
+    """Load HuggingFace model
+
+    Args:
+        model_name (str): Name of the model to load
+        trust_remote_code (bool): Whether to allow execution of code from the model repository.
+            Defaults to False for security. Only enable for trusted models.
+    """
+    if not HF_AVAILABLE:
+        raise ImportError("transformers not available")
+
+    if trust_remote_code:
+        print(
+            "⚠️  WARNING: Loading model with trust_remote_code=True. This can execute arbitrary code."
+        )
+
+    print(f"Loading HF: {model_name}")
+    tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=trust_remote_code)
+    model = AutoModelForCausalLM.from_pretrained(
+        model_name,
+        torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
+        device_map="auto",
+        trust_remote_code=trust_remote_code,
+    )
+    return tokenizer, model
+
+
+def load_vllm_model(model_name="Qwen/Qwen3-8B", trust_remote_code=False):
+    """Load vLLM model
+
+    Args:
+        model_name (str): Name of the model to load
+        trust_remote_code (bool): Whether to allow execution of code from the model repository.
+            Defaults to False for security. Only enable for trusted models.
+    """
+    if not VLLM_AVAILABLE:
+        raise ImportError("vllm not available")
+
+    if trust_remote_code:
+        print(
+            "⚠️  WARNING: Loading model with trust_remote_code=True. This can execute arbitrary code."
+        )
+
+    print(f"Loading vLLM: {model_name}")
+    llm = LLM(model=model_name, trust_remote_code=trust_remote_code)
+
+    # Qwen3 specific config
+    if is_qwen3_model(model_name):
+        stop_tokens = ["<|im_end|>", "<|end_of_text|>"]
+        max_tokens = 2048
+    else:
+        stop_tokens = None
+        max_tokens = 1024
+
+    sampling_params = SamplingParams(temperature=0.7, max_tokens=max_tokens, stop=stop_tokens)
+    return llm, sampling_params
+
+
+def generate_hf(tokenizer, model, prompt, max_tokens=None):
+    """Generate with HF - supports Qwen3 thinking models"""
+    model_name = getattr(model, "name_or_path", "unknown")
+    is_qwen3 = is_qwen3_model(model_name)
+
+    # Apply chat template for Qwen3
+    if is_qwen3:
+        prompt = apply_qwen3_chat_template(tokenizer, prompt)
+        max_tokens = max_tokens or 2048
+    else:
+        max_tokens = max_tokens or 1024
+
+    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
+    with torch.no_grad():
+        outputs = model.generate(
+            **inputs,
+            max_new_tokens=max_tokens,
+            temperature=0.7,
+            do_sample=True,
+            pad_token_id=tokenizer.eos_token_id,
+        )
+    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    response = response[len(prompt) :].strip()
+
+    # Extract final answer for thinking models
+    if is_qwen3:
+        return extract_thinking_answer(response)
+    return response
+
+
+def generate_vllm(llm, sampling_params, prompt):
+    """Generate with vLLM - supports Qwen3 thinking models"""
+    outputs = llm.generate([prompt], sampling_params)
+    response = outputs[0].outputs[0].text.strip()
+
+    # Extract final answer for Qwen3 thinking models
+    model_name = str(llm.llm_engine.model_config.model)
+    if is_qwen3_model(model_name):
+        return extract_thinking_answer(response)
+    return response
+
+
+def create_prompt(context, query, domain="default"):
+    """Create RAG prompt"""
+    if domain == "emails":
+        return f"Email content:\n{context}\n\nQuestion: {query}\n\nAnswer:"
+    elif domain == "finance":
+        return f"Financial content:\n{context}\n\nQuestion: {query}\n\nAnswer:"
+    elif domain == "multimodal":
+        return f"Image context:\n{context}\n\nQuestion: {query}\n\nAnswer:"
+    else:
+        return f"Context: {context}\n\nQuestion: {query}\n\nAnswer:"
+
+
+def evaluate_rag(searcher, llm_func, queries, domain="default", top_k=3, complexity=64):
+    """Simple RAG evaluation with timing"""
+    search_times = []
+    gen_times = []
+    results = []
+
+    for i, query in enumerate(queries):
+        # Search
+        start = time.time()
+        docs = searcher.search(query, top_k=top_k, complexity=complexity)
+        search_time = time.time() - start
+
+        # Generate
+        context = "\n\n".join([doc.text for doc in docs])
+        prompt = create_prompt(context, query, domain)
+
+        start = time.time()
+        response = llm_func(prompt)
+        gen_time = time.time() - start
+
+        search_times.append(search_time)
+        gen_times.append(gen_time)
+        results.append(response)
+
+        if i < 3:
+            print(f"Q{i + 1}: Search={search_time:.3f}s, Gen={gen_time:.3f}s")
+
+    return {
+        "avg_search_time": sum(search_times) / len(search_times),
+        "avg_generation_time": sum(gen_times) / len(gen_times),
+        "results": results,
+    }
+
+
+def load_qwen_vl_model(model_name="Qwen/Qwen2.5-VL-7B-Instruct", trust_remote_code=False):
+    """Load Qwen2.5-VL multimodal model
+
+    Args:
+        model_name (str): Name of the model to load
+        trust_remote_code (bool): Whether to allow execution of code from the model repository.
+            Defaults to False for security. Only enable for trusted models.
+    """
+    if not HF_AVAILABLE:
+        raise ImportError("transformers not available")
+
+    if trust_remote_code:
+        print(
+            "⚠️  WARNING: Loading model with trust_remote_code=True. This can execute arbitrary code."
+        )
+
+    print(f"Loading Qwen2.5-VL: {model_name}")
+
+    try:
+        from transformers import AutoModelForVision2Seq, AutoProcessor
+
+        processor = AutoProcessor.from_pretrained(model_name, trust_remote_code=trust_remote_code)
+        model = AutoModelForVision2Seq.from_pretrained(
+            model_name,
+            torch_dtype=torch.bfloat16,
+            device_map="auto",
+            trust_remote_code=trust_remote_code,
+        )
+
+        return processor, model
+
+    except Exception as e:
+        print(f"Failed to load with AutoModelForVision2Seq, trying specific class: {e}")
+
+        # Fallback to specific class
+        try:
+            from transformers import AutoProcessor, Qwen2VLForConditionalGeneration
+
+            processor = AutoProcessor.from_pretrained(
+                model_name, trust_remote_code=trust_remote_code
+            )
+            model = Qwen2VLForConditionalGeneration.from_pretrained(
+                model_name,
+                torch_dtype=torch.bfloat16,
+                device_map="auto",
+                trust_remote_code=trust_remote_code,
+            )
+
+            return processor, model
+
+        except Exception as e2:
+            raise ImportError(f"Failed to load Qwen2.5-VL model: {e2}")
+
+
+def generate_qwen_vl(processor, model, prompt, image_path=None, max_tokens=512):
+    """Generate with Qwen2.5-VL multimodal model"""
+    from PIL import Image
+
+    # Prepare inputs
+    if image_path:
+        image = Image.open(image_path)
+        inputs = processor(text=prompt, images=image, return_tensors="pt").to(model.device)
+    else:
+        inputs = processor(text=prompt, return_tensors="pt").to(model.device)
+
+    # Generate
+    with torch.no_grad():
+        generated_ids = model.generate(
+            **inputs, max_new_tokens=max_tokens, do_sample=False, temperature=0.1
+        )
+
+    # Decode response
+    generated_ids = generated_ids[:, inputs["input_ids"].shape[1] :]
+    response = processor.decode(generated_ids[0], skip_special_tokens=True)
+
+    return response
+
+
+def create_multimodal_prompt(context, query, image_descriptions, task_type="images"):
+    """Create prompt for multimodal RAG"""
+    if task_type == "images":
+        return f"""Based on the retrieved images and their descriptions, answer the following question.
+
+Retrieved Image Descriptions:
+{context}
+
+Question: {query}
+
+Provide a detailed answer based on the visual content described above."""
+
+    return f"Context: {context}\nQuestion: {query}\nAnswer:"
+
+
+def evaluate_multimodal_rag(searcher, queries, processor=None, model=None, complexity=64):
+    """Evaluate multimodal RAG with Qwen2.5-VL"""
+    search_times = []
+    gen_times = []
+    results = []
+
+    for i, query_item in enumerate(queries):
+        # Handle both string and dict formats for queries
+        if isinstance(query_item, dict):
+            query = query_item.get("query", "")
+            image_path = query_item.get("image_path")  # Optional reference image
+        else:
+            query = str(query_item)
+            image_path = None
+
+        # Search
+        start_time = time.time()
+        search_results = searcher.search(query, top_k=3, complexity=complexity)
+        search_time = time.time() - start_time
+        search_times.append(search_time)
+
+        # Prepare context from search results
+        context_parts = []
+        for result in search_results:
+            context_parts.append(f"- {result.text}")
+        context = "\n".join(context_parts)
+
+        # Generate with multimodal model
+        start_time = time.time()
+        if processor and model:
+            prompt = create_multimodal_prompt(context, query, context_parts)
+            response = generate_qwen_vl(processor, model, prompt, image_path)
+        else:
+            response = f"Context: {context}"
+        gen_time = time.time() - start_time
+
+        gen_times.append(gen_time)
+        results.append(response)
+
+        if i < 3:
+            print(f"Q{i + 1}: Search={search_time:.3f}s, Gen={gen_time:.3f}s")
+
+    return {
+        "avg_search_time": sum(search_times) / len(search_times),
+        "avg_generation_time": sum(gen_times) / len(gen_times),
+        "results": results,
+    }

benchmarks/run_all.sh

@@ -1,49 +0,0 @@
-#!/usr/bin/env bash
-set -euo pipefail
-
-# 公共参数
-INDEX_PATH="benchmarks/data/indices/rpj_wiki/rpj_wiki"
-NUM_QUERIES=20
-BATCH_SIZE=128
-LLM_MODEL="qwen3:4b"
-TOP_K=3
-
-# 日志目录（带时间戳）
-LOG_DIR="logs/eval_runs_$(date +%Y%m%d_%H%M%S)"
-mkdir -p "$LOG_DIR"
-
-# dataset -> ef 列表
-declare -A EF_MAP=(
-  [nq_open.jsonl]="32 62 190"
-  [trivia_qa.jsonl]="77 150 249"
-  [gpqa.jsonl]="41 72 124"
-  [hotpot_qa.jsonl]="137 299 1199"
-)
-
-# 按指定顺序遍历
-ORDERED_DATASETS=(nq_open.jsonl trivia_qa.jsonl gpqa.jsonl hotpot_qa.jsonl)
-
-for dataset in "${ORDERED_DATASETS[@]}"; do
-  for ef in ${EF_MAP[$dataset]}; do
-    log_file="${LOG_DIR}/${dataset%.jsonl}_ef${ef}.log"
-
-    # 展示并记录将要执行的命令
-    cmd=(python benchmarks/run_evaluation.py "$INDEX_PATH" \
-      --num-queries "$NUM_QUERIES" \
-      --ef "$ef" \
-      --batch-size "$BATCH_SIZE" \
-      --llm-model "$LLM_MODEL" \
-      --top-k "$TOP_K" \
-      --queries-file "$dataset")
-
-    echo "=== Running dataset=${dataset} ef=${ef} ===" | tee -a "$log_file"
-    printf 'CMD: '; printf '%q ' "${cmd[@]}" | tee -a "$log_file"; echo | tee -a "$log_file"
-
-    # 同时输出到命令行和日志文件
-    "${cmd[@]}" 2>&1 | tee -a "$log_file"
-
-    echo | tee -a "$log_file"
-  done
-done
-
-echo "All runs completed. Logs in: $LOG_DIR"

benchmarks/run_evaluation.py

@@ -53,7 +53,7 @@ def download_data_if_needed(data_root: Path, download_embeddings: bool = False):
             print(
                 "Error: huggingface_hub is not installed. Please install it to download the data:"
             )
-            print("uv pip install -e '.[dev]'")
+            print("uv sync --only-group dev")
             sys.exit(1)
         except Exception as e:
             print(f"An error occurred during data download: {e}")
@@ -203,15 +203,6 @@ def main():
         default=0,
         help="Batch size for HNSW batched search (0 disables batching)",
     )
-    parser.add_argument(
-        "--queries-file",
-        type=str,
-        default="nq_open.jsonl",
-        help=(
-            "Queries file to use. Provide a filename under benchmarks/data/queries "
-            "or an absolute path to a .jsonl file (default: nq_open.jsonl)."
-        ),
-    )
     parser.add_argument(
         "--llm-type",
         type=str,
@@ -323,52 +314,8 @@ def main():
         dataset_type = Path(args.index_path).name
         print(f"WARNING: Could not detect dataset type from path, inferred '{dataset_type}'.")
 
-    # Resolve queries file (supports absolute path or name under data/queries)
-    queries_file_candidate = Path(args.queries_file)
-    if queries_file_candidate.is_absolute():
-        queries_file = queries_file_candidate
-    else:
-        queries_file = data_root / "queries" / args.queries_file
-
-    if not queries_file.exists():
-        print(f"Error: Queries file not found: {queries_file}")
-        print("Tip: Use --queries-file with a filename under benchmarks/data/queries or an absolute path.")
-        sys.exit(1)
-
-    # Infer ground-truth file from the queries filename
-    qname = queries_file.name.lower()
-    if "hotpot" in qname:
-        task_key = "hotpot"
-    elif "trivia" in qname:
-        task_key = "trivia"
-    elif "gpqa" in qname:
-        task_key = "gpqa"
-    elif "nq" in qname:
-        task_key = "nq"
-    else:
-        print(
-            "Error: Could not infer task from queries filename. Supported names include 'nq', 'hotpot', 'trivia', 'gpqa'."
-        )
-        print(f"Filename was: {queries_file.name}")
-        sys.exit(1)
-
-    golden_results_file = data_root / "ground_truth" / dataset_type / f"flat_results_{task_key}_k3.json"
-    if not golden_results_file.exists():
-        gt_dir = data_root / "ground_truth" / dataset_type
-        try:
-            available = sorted(p.name for p in gt_dir.glob("flat_results_*_k3.json"))
-        except Exception:
-            available = []
-        print(
-            f"Error: Ground truth file not found for task '{task_key}' under dataset '{dataset_type}': {golden_results_file}"
-        )
-        if available:
-            print("Available ground truth files:")
-            for name in available:
-                print(f"  - {name}")
-        else:
-            print(f"No ground truth files found in {gt_dir}")
-        sys.exit(1)
+    queries_file = data_root / "queries" / "nq_open.jsonl"
+    golden_results_file = data_root / "ground_truth" / dataset_type / "flat_results_nq_k3.json"
 
     print(f"INFO: Detected dataset type: {dataset_type}")
     print(f"INFO: Using queries file: {queries_file}")
@@ -399,15 +346,15 @@ def main():
             search_times.append(time.time() - start_time)
 
             # Optional: also call the LLM with configurable backend/model (does not affect recall)
-            # llm_config = {"type": args.llm_type, "model": args.llm_model}
-            # chat = LeannChat(args.index_path, llm_config=llm_config, searcher=searcher)
-            # answer = chat.ask(
-            #     queries[i],
-            #     top_k=args.top_k,
-            #     complexity=args.ef_search,
-            #     batch_size=args.batch_size,
-            # )
-            # print(f"Answer: {answer}")
+            llm_config = {"type": args.llm_type, "model": args.llm_model}
+            chat = LeannChat(args.index_path, llm_config=llm_config, searcher=searcher)
+            answer = chat.ask(
+                queries[i],
+                top_k=args.top_k,
+                complexity=args.ef_search,
+                batch_size=args.batch_size,
+            )
+            print(f"Answer: {answer}")
             # Correct Recall Calculation: Based on TEXT content
             new_texts = {result.text for result in new_results}
 
@@ -431,16 +378,10 @@ def main():
         avg_recall = np.mean(recall_scores) if recall_scores else 0
         avg_time = np.mean(search_times) if search_times else 0
 
-        print(f"search time: {search_times}")
-
         print("\n🎉 --- Evaluation Complete ---")
         print(f"Avg. Recall@{args.top_k} (efSearch={args.ef_search}): {avg_recall:.4f}")
         print(f"Avg. Search Time: {avg_time:.4f}s")
 
-        # avg last 10 search times
-        avg_last_10_search_times = np.mean(search_times[-10:])
-        print(f"Avg. Last 10 Search Times: {avg_last_10_search_times:.4f}s")
-
     except Exception as e:
         print(f"\n❌ An error occurred during evaluation: {e}")
         import traceback

benchmarks/run_speed_bench_all.sh

@@ -1,55 +0,0 @@
-#!/usr/bin/env bash
-set -euo pipefail
-
-# Absolute paths (adjust if needed)
-PROMPTS_DIR="/home/tony/yichuan/leann/benchmarks/data/prompts_g5"
-SCRIPT_PATH="/home/tony/yichuan/leann/benchmarks/generation_speed_bench.py"
-
-# Common args
-MAX_TOKENS=2048
-OLLAMA_MODEL="qwen3:4b"
-HF_MODEL="Qwen/Qwen3-4B"
-
-# Logs
-LOG_DIR="/home/tony/yichuan/leann/logs/speed_bench_$(date +%Y%m%d_%H%M%S)"
-mkdir -p "$LOG_DIR"
-
-echo "Scanning: $PROMPTS_DIR"
-
-# Iterate all .txt files under PROMPTS_DIR
-while IFS= read -r -d '' file; do
-  base_name=$(basename "$file")
-  stem_name="${base_name%.*}"
-
-  # 1) Ollama
-  log_ollama="${LOG_DIR}/${stem_name}_ollama.log"
-  cmd_ollama=(python "$SCRIPT_PATH" \
-    --path "$file" \
-    --type ollama \
-    --model "$OLLAMA_MODEL" \
-    --max_tokens "$MAX_TOKENS")
-
-  echo "=== Running (ollama) file=${file} model=${OLLAMA_MODEL} ===" | tee -a "$log_ollama"
-  printf 'CMD: '; printf '%q ' "${cmd_ollama[@]}" | tee -a "$log_ollama"; echo | tee -a "$log_ollama"
-  "${cmd_ollama[@]}" 2>&1 | tee -a "$log_ollama"
-  echo | tee -a "$log_ollama"
-
-  # 2) HF
-  log_hf="${LOG_DIR}/${stem_name}_hf.log"
-  cmd_hf=(python "$SCRIPT_PATH" \
-    --path "$file" \
-    --type hf \
-    --model "$HF_MODEL" \
-    --max_tokens "$MAX_TOKENS")
-
-  echo "=== Running (hf) file=${file} model=${HF_MODEL} ===" | tee -a "$log_hf"
-  printf 'CMD: '; printf '%q ' "${cmd_hf[@]}" | tee -a "$log_hf"; echo | tee -a "$log_hf"
-  "${cmd_hf[@]}" 2>&1 | tee -a "$log_hf"
-  echo | tee -a "$log_hf"
-
-done < <(find "$PROMPTS_DIR" -type f -name '*.txt' -print0)
-
-
-echo "All runs completed. Logs in: $LOG_DIR"
-
-

docs/CONTRIBUTING.md

@@ -53,9 +53,9 @@ We use pre-commit hooks to ensure code quality and consistency. This runs automa
 
 ### Setup Pre-commit
 
-1. **Install pre-commit** (already included when you run `uv sync`):
+1. **Install pre-commit tools**:
    ```bash
-   uv pip install pre-commit
+   uv sync lint
    ```
 
 2. **Install the git hooks**:
@@ -65,7 +65,7 @@ We use pre-commit hooks to ensure code quality and consistency. This runs automa
 
 3. **Run pre-commit manually** (optional):
    ```bash
-   pre-commit run --all-files
+   uv run pre-commit run --all-files
    ```
 
 ### Pre-commit Checks
@@ -85,6 +85,9 @@ Our pre-commit configuration includes:
 ### Running Tests
 
 ```bash
+# Install test tools only (no project runtime)
+uv sync --group test
+
 # Run all tests
 uv run pytest
 

docs/ast_chunking_guide.md

@@ -26,6 +26,21 @@ leann build my-code-index --docs ./src --use-ast-chunking
 uv pip install -e "."
 ```
 
+#### For normal users (PyPI install)
+- Use `pip install leann` or `uv pip install leann`.
+- `astchunk` is pulled automatically from PyPI as a dependency; no extra steps.
+
+#### For developers (from source, editable)
+```bash
+git clone https://github.com/yichuan-w/LEANN.git leann
+cd leann
+git submodule update --init --recursive
+uv sync
+```
+- This repo vendors `astchunk` as a git submodule at `packages/astchunk-leann` (our fork).
+- `[tool.uv.sources]` maps the `astchunk` package to that path in editable mode.
+- You can edit code under `packages/astchunk-leann` and Python will use your changes immediately (no separate `pip install astchunk` needed).
+
 ## Best Practices
 
 ### When to Use AST Chunking

docs/configuration-guide.md

@@ -83,6 +83,81 @@ ollama pull nomic-embed-text
 
 </details>
 
+## Local & Remote Inference Endpoints
+
+> Applies to both LLMs (`leann ask`) and embeddings (`leann build`).
+
+LEANN now treats Ollama, LM Studio, and other OpenAI-compatible runtimes as first-class providers. You can point LEANN at any compatible endpoint – either on the same machine or across the network – with a couple of flags or environment variables.
+
+### One-Time Environment Setup
+
+```bash
+# Works for OpenAI-compatible runtimes such as LM Studio, vLLM, SGLang, llamafile, etc.
+export OPENAI_API_KEY="your-key"            # or leave unset for local servers that do not check keys
+export OPENAI_BASE_URL="http://localhost:1234/v1"
+
+# Ollama-compatible runtimes (Ollama, Ollama on another host, llamacpp-server, etc.)
+export LEANN_OLLAMA_HOST="http://localhost:11434"   # falls back to OLLAMA_HOST or LOCAL_LLM_ENDPOINT
+```
+
+LEANN also recognises `LEANN_LOCAL_LLM_HOST` (highest priority), `LEANN_OPENAI_BASE_URL`, and `LOCAL_OPENAI_BASE_URL`, so existing scripts continue to work.
+
+### Passing Hosts Per Command
+
+```bash
+# Build an index with a remote embedding server
+leann build my-notes \
+  --docs ./notes \
+  --embedding-mode openai \
+  --embedding-model text-embedding-qwen3-embedding-0.6b \
+  --embedding-api-base http://192.168.1.50:1234/v1 \
+  --embedding-api-key local-dev-key
+
+# Query using a local LM Studio instance via OpenAI-compatible API
+leann ask my-notes \
+  --llm openai \
+  --llm-model qwen3-8b \
+  --api-base http://localhost:1234/v1 \
+  --api-key local-dev-key
+
+# Query an Ollama instance running on another box
+leann ask my-notes \
+  --llm ollama \
+  --llm-model qwen3:14b \
+  --host http://192.168.1.101:11434
+```
+
+⚠️ **Make sure the endpoint is reachable**: when your inference server runs on a home/workstation and the index/search job runs in the cloud, the server must be able to reach the host you configured. Typical options include:
+
+- Expose a public IP (and open the relevant port) on the machine that hosts LM Studio/Ollama.
+- Configure router or cloud provider port forwarding.
+- Tunnel traffic through tools like `tailscale`, `cloudflared`, or `ssh -R`.
+
+When you set these options while building an index, LEANN stores them in `meta.json`. Any subsequent `leann ask` or searcher process automatically reuses the same provider settings – even when we spawn background embedding servers. This makes the “server without GPU talking to my local workstation” workflow from [issue #80](https://github.com/yichuan-w/LEANN/issues/80#issuecomment-2287230548) work out-of-the-box.
+
+**Tip:** If your runtime does not require an API key (many local stacks don’t), leave `--api-key` unset. LEANN will skip injecting credentials.
+
+### Python API Usage
+
+You can pass the same configuration from Python:
+
+```python
+from leann.api import LeannBuilder
+
+builder = LeannBuilder(
+    backend_name="hnsw",
+    embedding_mode="openai",
+    embedding_model="text-embedding-qwen3-embedding-0.6b",
+    embedding_options={
+        "base_url": "http://192.168.1.50:1234/v1",
+        "api_key": "local-dev-key",
+    },
+)
+builder.build_index("./indexes/my-notes", chunks)
+```
+
+`embedding_options` is persisted to the index `meta.json`, so subsequent `LeannSearcher` or `LeannChat` sessions automatically reuse the same provider settings (the embedding server manager forwards them to the provider for you).
+
 ## Index Selection: Matching Your Scale
 
 ### HNSW (Hierarchical Navigable Small World)
@@ -380,5 +455,5 @@ Conclusion:
 
 - [Lessons Learned Developing LEANN](https://yichuan-w.github.io/blog/lessons_learned_in_dev_leann/)
 - [LEANN Technical Paper](https://arxiv.org/abs/2506.08276)
-- [DiskANN Original Paper](https://papers.nips.cc/paper/2019/file/09853c7fb1d3f8ee67a61b6bf4a7f8e6-Paper.pdf)
+- [DiskANN Original Paper](https://suhasjs.github.io/files/diskann_neurips19.pdf)
 - [SSD-based Graph Partitioning](https://github.com/SonglinLife/SSD_BASED_PLAN)

docs/grep_search.md

@@ -0,0 +1,149 @@
+# LEANN Grep Search Usage Guide
+
+## Overview
+
+LEANN's grep search functionality provides exact text matching for finding specific code patterns, error messages, function names, or exact phrases in your indexed documents.
+
+## Basic Usage
+
+### Simple Grep Search
+
+```python
+from leann.api import LeannSearcher
+
+searcher = LeannSearcher("your_index_path")
+
+# Exact text search
+results = searcher.search("def authenticate_user", use_grep=True, top_k=5)
+
+for result in results:
+    print(f"Score: {result.score}")
+    print(f"Text: {result.text[:100]}...")
+    print("-" * 40)
+```
+
+### Comparison: Semantic vs Grep Search
+
+```python
+# Semantic search - finds conceptually similar content
+semantic_results = searcher.search("machine learning algorithms", top_k=3)
+
+# Grep search - finds exact text matches
+grep_results = searcher.search("def train_model", use_grep=True, top_k=3)
+```
+
+## When to Use Grep Search
+
+### Use Cases
+
+- **Code Search**: Finding specific function definitions, class names, or variable references
+- **Error Debugging**: Locating exact error messages or stack traces
+- **Documentation**: Finding specific API endpoints or exact terminology
+
+### Examples
+
+```python
+# Find function definitions
+functions = searcher.search("def __init__", use_grep=True)
+
+# Find import statements
+imports = searcher.search("from sklearn import", use_grep=True)
+
+# Find specific error types
+errors = searcher.search("FileNotFoundError", use_grep=True)
+
+# Find TODO comments
+todos = searcher.search("TODO:", use_grep=True)
+
+# Find configuration entries
+configs = searcher.search("server_port=", use_grep=True)
+```
+
+## Technical Details
+
+### How It Works
+
+1. **File Location**: Grep search operates on the raw text stored in `.jsonl` files
+2. **Command Execution**: Uses the system `grep` command with case-insensitive search
+3. **Result Processing**: Parses JSON lines and extracts text and metadata
+4. **Scoring**: Simple frequency-based scoring based on query term occurrences
+
+### Search Process
+
+```
+Query: "def train_model"
+  ↓
+grep -i -n "def train_model" documents.leann.passages.jsonl
+  ↓
+Parse matching JSON lines
+  ↓
+Calculate scores based on term frequency
+  ↓
+Return top_k results
+```
+
+### Scoring Algorithm
+
+```python
+# Term frequency in document
+score = text.lower().count(query.lower())
+```
+
+Results are ranked by score (highest first), with higher scores indicating more occurrences of the search term.
+
+## Error Handling
+
+### Common Issues
+
+#### Grep Command Not Found
+```
+RuntimeError: grep command not found. Please install grep or use semantic search.
+```
+
+**Solution**: Install grep on your system:
+- **Ubuntu/Debian**: `sudo apt-get install grep`
+- **macOS**: grep is pre-installed
+- **Windows**: Use WSL or install grep via Git Bash/MSYS2
+
+#### No Results Found
+```python
+# Check if your query exists in the raw data
+results = searcher.search("your_query", use_grep=True)
+if not results:
+    print("No exact matches found. Try:")
+    print("1. Check spelling and case")
+    print("2. Use partial terms")
+    print("3. Switch to semantic search")
+```
+
+## Complete Example
+
+```python
+#!/usr/bin/env python3
+"""
+Grep Search Example
+Demonstrates grep search for exact text matching.
+"""
+
+from leann.api import LeannSearcher
+
+def demonstrate_grep_search():
+    # Initialize searcher
+    searcher = LeannSearcher("my_index")
+
+    print("=== Function Search ===")
+    functions = searcher.search("def __init__", use_grep=True, top_k=5)
+    for i, result in enumerate(functions, 1):
+        print(f"{i}. Score: {result.score}")
+        print(f"   Preview: {result.text[:60]}...")
+        print()
+
+    print("=== Error Search ===")
+    errors = searcher.search("FileNotFoundError", use_grep=True, top_k=3)
+    for result in errors:
+        print(f"Content: {result.text.strip()}")
+        print("-" * 40)
+
+if __name__ == "__main__":
+    demonstrate_grep_search()
+```

examples/dynamic_update_no_recompute.py

@@ -0,0 +1,429 @@
+"""Dynamic HNSW update demo without compact storage.
+
+This script reproduces the minimal scenario we used while debugging on-the-fly
+recompute:
+
+1. Build a non-compact HNSW index from the first few paragraphs of a text file.
+2. Print the top results with `recompute_embeddings=True`.
+3. Append additional paragraphs with :meth:`LeannBuilder.update_index`.
+4. Run the same query again to show the newly inserted passages.
+
+Run it with ``uv`` (optionally pointing LEANN_HNSW_LOG_PATH at a file to inspect
+ZMQ activity)::
+
+    LEANN_HNSW_LOG_PATH=embedding_fetch.log \
+    uv run -m examples.dynamic_update_no_recompute \
+      --index-path .leann/examples/leann-demo.leann
+
+By default the script builds an index from ``data/2501.14312v1 (1).pdf`` and
+then updates it with LEANN-related material from ``data/2506.08276v1.pdf``.
+It issues the query "What's LEANN?" before and after the update to show how the
+new passages become immediately searchable. The script uses the
+``sentence-transformers/all-MiniLM-L6-v2`` model with ``is_recompute=True`` so
+Faiss pulls existing vectors on demand via the ZMQ embedding server, while
+freshly added passages are embedded locally just like the initial build.
+
+To make storage comparisons easy, the script can also build a matching
+``is_recompute=False`` baseline (enabled by default) and report the index size
+delta after the update. Disable the baseline run with
+``--skip-compare-no-recompute`` if you only need the recompute flow.
+"""
+
+import argparse
+import json
+from collections.abc import Iterable
+from pathlib import Path
+from typing import Any
+
+from leann.api import LeannBuilder, LeannSearcher
+from leann.registry import register_project_directory
+
+from apps.chunking import create_text_chunks
+
+REPO_ROOT = Path(__file__).resolve().parents[1]
+
+DEFAULT_QUERY = "What's LEANN?"
+DEFAULT_INITIAL_FILES = [
+    REPO_ROOT / "data" / "2501.14312v1 (1).pdf",
+    REPO_ROOT / "data" / "huawei_pangu.md",
+    REPO_ROOT / "data" / "PrideandPrejudice.txt",
+]
+DEFAULT_UPDATE_FILES = [REPO_ROOT / "data" / "2506.08276v1.pdf"]
+
+
+def load_chunks_from_files(paths: list[Path]) -> list[str]:
+    from llama_index.core import SimpleDirectoryReader
+
+    documents = []
+    for path in paths:
+        p = path.expanduser().resolve()
+        if not p.exists():
+            raise FileNotFoundError(f"Input path not found: {p}")
+        if p.is_dir():
+            reader = SimpleDirectoryReader(str(p), recursive=False)
+            documents.extend(reader.load_data(show_progress=True))
+        else:
+            reader = SimpleDirectoryReader(input_files=[str(p)])
+            documents.extend(reader.load_data(show_progress=True))
+
+    if not documents:
+        return []
+
+    chunks = create_text_chunks(
+        documents,
+        chunk_size=512,
+        chunk_overlap=128,
+        use_ast_chunking=False,
+    )
+    return [c for c in chunks if isinstance(c, str) and c.strip()]
+
+
+def run_search(index_path: Path, query: str, top_k: int, *, recompute_embeddings: bool) -> list:
+    searcher = LeannSearcher(str(index_path))
+    try:
+        return searcher.search(
+            query=query,
+            top_k=top_k,
+            recompute_embeddings=recompute_embeddings,
+            batch_size=16,
+        )
+    finally:
+        searcher.cleanup()
+
+
+def print_results(title: str, results: Iterable) -> None:
+    print(f"\n=== {title} ===")
+    res_list = list(results)
+    print(f"results count: {len(res_list)}")
+    print("passages:")
+    if not res_list:
+        print("  (no passages returned)")
+    for res in res_list:
+        snippet = res.text.replace("\n", " ")[:120]
+        print(f"  - {res.id}: {snippet}... (score={res.score:.4f})")
+
+
+def build_initial_index(
+    index_path: Path,
+    paragraphs: list[str],
+    model_name: str,
+    embedding_mode: str,
+    is_recompute: bool,
+) -> None:
+    builder = LeannBuilder(
+        backend_name="hnsw",
+        embedding_model=model_name,
+        embedding_mode=embedding_mode,
+        is_compact=False,
+        is_recompute=is_recompute,
+    )
+    for idx, passage in enumerate(paragraphs):
+        builder.add_text(passage, metadata={"id": str(idx)})
+    builder.build_index(str(index_path))
+
+
+def update_index(
+    index_path: Path,
+    start_id: int,
+    paragraphs: list[str],
+    model_name: str,
+    embedding_mode: str,
+    is_recompute: bool,
+) -> None:
+    updater = LeannBuilder(
+        backend_name="hnsw",
+        embedding_model=model_name,
+        embedding_mode=embedding_mode,
+        is_compact=False,
+        is_recompute=is_recompute,
+    )
+    for offset, passage in enumerate(paragraphs, start=start_id):
+        updater.add_text(passage, metadata={"id": str(offset)})
+    updater.update_index(str(index_path))
+
+
+def ensure_index_dir(index_path: Path) -> None:
+    index_path.parent.mkdir(parents=True, exist_ok=True)
+
+
+def cleanup_index_files(index_path: Path) -> None:
+    """Remove leftover index artifacts for a clean rebuild."""
+
+    parent = index_path.parent
+    if not parent.exists():
+        return
+    stem = index_path.stem
+    for file in parent.glob(f"{stem}*"):
+        if file.is_file():
+            file.unlink()
+
+
+def index_file_size(index_path: Path) -> int:
+    """Return the size of the primary .index file for the given index path."""
+
+    index_file = index_path.parent / f"{index_path.stem}.index"
+    return index_file.stat().st_size if index_file.exists() else 0
+
+
+def load_metadata_snapshot(index_path: Path) -> dict[str, Any] | None:
+    meta_path = index_path.parent / f"{index_path.name}.meta.json"
+    if not meta_path.exists():
+        return None
+    try:
+        return json.loads(meta_path.read_text())
+    except json.JSONDecodeError:
+        return None
+
+
+def run_workflow(
+    *,
+    label: str,
+    index_path: Path,
+    initial_paragraphs: list[str],
+    update_paragraphs: list[str],
+    model_name: str,
+    embedding_mode: str,
+    is_recompute: bool,
+    query: str,
+    top_k: int,
+    skip_search: bool,
+) -> dict[str, Any]:
+    prefix = f"[{label}] " if label else ""
+
+    ensure_index_dir(index_path)
+    cleanup_index_files(index_path)
+
+    print(f"{prefix}Building initial index...")
+    build_initial_index(
+        index_path,
+        initial_paragraphs,
+        model_name,
+        embedding_mode,
+        is_recompute=is_recompute,
+    )
+
+    initial_size = index_file_size(index_path)
+    if not skip_search:
+        before_results = run_search(
+            index_path,
+            query,
+            top_k,
+            recompute_embeddings=is_recompute,
+        )
+    else:
+        before_results = None
+
+    print(f"\n{prefix}Updating index with additional passages...")
+    update_index(
+        index_path,
+        start_id=len(initial_paragraphs),
+        paragraphs=update_paragraphs,
+        model_name=model_name,
+        embedding_mode=embedding_mode,
+        is_recompute=is_recompute,
+    )
+
+    if not skip_search:
+        after_results = run_search(
+            index_path,
+            query,
+            top_k,
+            recompute_embeddings=is_recompute,
+        )
+    else:
+        after_results = None
+    updated_size = index_file_size(index_path)
+
+    return {
+        "initial_size": initial_size,
+        "updated_size": updated_size,
+        "delta": updated_size - initial_size,
+        "before_results": before_results if not skip_search else None,
+        "after_results": after_results if not skip_search else None,
+        "metadata": load_metadata_snapshot(index_path),
+    }
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument(
+        "--initial-files",
+        type=Path,
+        nargs="+",
+        default=DEFAULT_INITIAL_FILES,
+        help="Initial document files (PDF/TXT) used to build the base index",
+    )
+    parser.add_argument(
+        "--index-path",
+        type=Path,
+        default=Path(".leann/examples/leann-demo.leann"),
+        help="Destination index path (default: .leann/examples/leann-demo.leann)",
+    )
+    parser.add_argument(
+        "--initial-count",
+        type=int,
+        default=8,
+        help="Number of chunks to use from the initial documents (default: 8)",
+    )
+    parser.add_argument(
+        "--update-files",
+        type=Path,
+        nargs="*",
+        default=DEFAULT_UPDATE_FILES,
+        help="Additional documents to add during update (PDF/TXT)",
+    )
+    parser.add_argument(
+        "--update-count",
+        type=int,
+        default=4,
+        help="Number of chunks to append from update documents (default: 4)",
+    )
+    parser.add_argument(
+        "--update-text",
+        type=str,
+        default=(
+            "LEANN (Lightweight Embedding ANN) is an indexing toolkit focused on "
+            "recompute-aware HNSW graphs, allowing embeddings to be regenerated "
+            "on demand to keep disk usage minimal."
+        ),
+        help="Fallback text to append if --update-files is omitted",
+    )
+    parser.add_argument(
+        "--top-k",
+        type=int,
+        default=4,
+        help="Number of results to show for each search (default: 4)",
+    )
+    parser.add_argument(
+        "--query",
+        type=str,
+        default=DEFAULT_QUERY,
+        help="Query to run before/after the update",
+    )
+    parser.add_argument(
+        "--embedding-model",
+        type=str,
+        default="sentence-transformers/all-MiniLM-L6-v2",
+        help="Embedding model name",
+    )
+    parser.add_argument(
+        "--embedding-mode",
+        type=str,
+        default="sentence-transformers",
+        choices=["sentence-transformers", "openai", "mlx", "ollama"],
+        help="Embedding backend mode",
+    )
+    parser.add_argument(
+        "--compare-no-recompute",
+        dest="compare_no_recompute",
+        action="store_true",
+        help="Also run a baseline with is_recompute=False and report its index growth.",
+    )
+    parser.add_argument(
+        "--skip-compare-no-recompute",
+        dest="compare_no_recompute",
+        action="store_false",
+        help="Skip building the no-recompute baseline.",
+    )
+    parser.add_argument(
+        "--skip-search",
+        dest="skip_search",
+        action="store_true",
+        help="Skip the search step.",
+    )
+    parser.set_defaults(compare_no_recompute=True)
+    args = parser.parse_args()
+
+    ensure_index_dir(args.index_path)
+    register_project_directory(REPO_ROOT)
+
+    initial_chunks = load_chunks_from_files(list(args.initial_files))
+    if not initial_chunks:
+        raise ValueError("No text chunks extracted from the initial files.")
+
+    initial = initial_chunks[: args.initial_count]
+    if not initial:
+        raise ValueError("Initial chunk set is empty after applying --initial-count.")
+
+    if args.update_files:
+        update_chunks = load_chunks_from_files(list(args.update_files))
+        if not update_chunks:
+            raise ValueError("No text chunks extracted from the update files.")
+        to_add = update_chunks[: args.update_count]
+    else:
+        if not args.update_text:
+            raise ValueError("Provide --update-files or --update-text for the update step.")
+        to_add = [args.update_text]
+    if not to_add:
+        raise ValueError("Update chunk set is empty after applying --update-count.")
+
+    recompute_stats = run_workflow(
+        label="recompute",
+        index_path=args.index_path,
+        initial_paragraphs=initial,
+        update_paragraphs=to_add,
+        model_name=args.embedding_model,
+        embedding_mode=args.embedding_mode,
+        is_recompute=True,
+        query=args.query,
+        top_k=args.top_k,
+        skip_search=args.skip_search,
+    )
+
+    if not args.skip_search:
+        print_results("initial search", recompute_stats["before_results"])
+    if not args.skip_search:
+        print_results("after update", recompute_stats["after_results"])
+    print(
+        f"\n[recompute] Index file size change: {recompute_stats['initial_size']} -> {recompute_stats['updated_size']} bytes"
+        f" (Δ {recompute_stats['delta']})"
+    )
+
+    if recompute_stats["metadata"]:
+        meta_view = {k: recompute_stats["metadata"].get(k) for k in ("is_compact", "is_pruned")}
+        print("[recompute] metadata snapshot:")
+        print(json.dumps(meta_view, indent=2))
+
+    if args.compare_no_recompute:
+        baseline_path = (
+            args.index_path.parent / f"{args.index_path.stem}-norecompute{args.index_path.suffix}"
+        )
+        baseline_stats = run_workflow(
+            label="no-recompute",
+            index_path=baseline_path,
+            initial_paragraphs=initial,
+            update_paragraphs=to_add,
+            model_name=args.embedding_model,
+            embedding_mode=args.embedding_mode,
+            is_recompute=False,
+            query=args.query,
+            top_k=args.top_k,
+            skip_search=args.skip_search,
+        )
+
+        print(
+            f"\n[no-recompute] Index file size change: {baseline_stats['initial_size']} -> {baseline_stats['updated_size']} bytes"
+            f" (Δ {baseline_stats['delta']})"
+        )
+
+        after_texts = (
+            [res.text for res in recompute_stats["after_results"]] if not args.skip_search else None
+        )
+        baseline_after_texts = (
+            [res.text for res in baseline_stats["after_results"]] if not args.skip_search else None
+        )
+        if after_texts == baseline_after_texts:
+            print(
+                "[no-recompute] Search results match recompute baseline; see above for the shared output."
+            )
+        else:
+            print("[no-recompute] WARNING: search results differ from recompute baseline.")
+
+        if baseline_stats["metadata"]:
+            meta_view = {k: baseline_stats["metadata"].get(k) for k in ("is_compact", "is_pruned")}
+            print("[no-recompute] metadata snapshot:")
+            print(json.dumps(meta_view, indent=2))
+
+
+if __name__ == "__main__":
+    main()

examples/grep_search_example.py

@@ -0,0 +1,35 @@
+"""
+Grep Search Example
+
+Shows how to use grep-based text search instead of semantic search.
+Useful when you need exact text matches rather than meaning-based results.
+"""
+
+from leann import LeannSearcher
+
+# Load your index
+searcher = LeannSearcher("my-documents.leann")
+
+# Regular semantic search
+print("=== Semantic Search ===")
+results = searcher.search("machine learning algorithms", top_k=3)
+for result in results:
+    print(f"Score: {result.score:.3f}")
+    print(f"Text: {result.text[:80]}...")
+    print()
+
+# Grep-based search for exact text matches
+print("=== Grep Search ===")
+results = searcher.search("def train_model", top_k=3, use_grep=True)
+for result in results:
+    print(f"Score: {result.score}")
+    print(f"Text: {result.text[:80]}...")
+    print()
+
+# Find specific error messages
+error_results = searcher.search("FileNotFoundError", use_grep=True)
+print(f"Found {len(error_results)} files mentioning FileNotFoundError")
+
+# Search for function definitions
+func_results = searcher.search("class SearchResult", use_grep=True, top_k=5)
+print(f"Found {len(func_results)} class definitions")

llms.txt

@@ -0,0 +1,28 @@
+# llms.txt — LEANN MCP and Agent Integration
+product: LEANN
+homepage: https://github.com/yichuan-w/LEANN
+contact: https://github.com/yichuan-w/LEANN/issues
+
+# Installation
+install: uv tool install leann-core --with leann
+
+# MCP Server Entry Point
+mcp.server: leann_mcp
+mcp.protocol_version: 2024-11-05
+
+# Tools
+mcp.tools: leann_list, leann_search
+
+mcp.tool.leann_list.description: List available LEANN indexes
+mcp.tool.leann_list.input: {}
+
+mcp.tool.leann_search.description: Semantic search across a named LEANN index
+mcp.tool.leann_search.input.index_name: string, required
+mcp.tool.leann_search.input.query: string, required
+mcp.tool.leann_search.input.top_k: integer, optional, default=5, min=1, max=20
+mcp.tool.leann_search.input.complexity: integer, optional, default=32, min=16, max=128
+
+# Notes
+note: Build indexes with `leann build <name> --docs <files...>` before searching.
+example.add: claude mcp add --scope user leann-server -- leann_mcp
+example.verify: claude mcp list | cat

packages/leann-backend-diskann/leann_backend_diskann/diskann_backend.py

@@ -343,7 +343,8 @@ class DiskannSearcher(BaseSearcher):
                 "full_index_prefix": full_index_prefix,
                 "num_threads": self.num_threads,
                 "num_nodes_to_cache": kwargs.get("num_nodes_to_cache", 0),
-                "cache_mechanism": 1,
+                # 1 -> initialize cache using sample_data; 2 -> ready cache without init; others disable cache
+                "cache_mechanism": kwargs.get("cache_mechanism", 1),
                 "pq_prefix": "",
                 "partition_prefix": partition_prefix,
             }

packages/leann-backend-diskann/leann_backend_diskann/diskann_embedding_server.py

@@ -10,7 +10,7 @@ import sys
 import threading
 import time
 from pathlib import Path
-from typing import Optional
+from typing import Any, Optional
 
 import numpy as np
 import zmq
@@ -32,6 +32,16 @@ if not logger.handlers:
     logger.propagate = False
 
 
+_RAW_PROVIDER_OPTIONS = os.getenv("LEANN_EMBEDDING_OPTIONS")
+try:
+    PROVIDER_OPTIONS: dict[str, Any] = (
+        json.loads(_RAW_PROVIDER_OPTIONS) if _RAW_PROVIDER_OPTIONS else {}
+    )
+except json.JSONDecodeError:
+    logger.warning("Failed to parse LEANN_EMBEDDING_OPTIONS; ignoring provider options")
+    PROVIDER_OPTIONS = {}
+
+
 def create_diskann_embedding_server(
     passages_file: Optional[str] = None,
     zmq_port: int = 5555,
@@ -181,7 +191,12 @@ def create_diskann_embedding_server(
                     logger.debug(f"Text lengths: {[len(t) for t in texts[:5]]}")  # Show first 5
 
                 # Process embeddings using unified computation
-                embeddings = compute_embeddings(texts, model_name, mode=embedding_mode)
+                embeddings = compute_embeddings(
+                    texts,
+                    model_name,
+                    mode=embedding_mode,
+                    provider_options=PROVIDER_OPTIONS,
+                )
                 logger.info(
                     f"Computed embeddings for {len(texts)} texts, shape: {embeddings.shape}"
                 )
@@ -296,7 +311,12 @@ def create_diskann_embedding_server(
                             continue
 
                     # Process the request
-                    embeddings = compute_embeddings(texts, model_name, mode=embedding_mode)
+                    embeddings = compute_embeddings(
+                        texts,
+                        model_name,
+                        mode=embedding_mode,
+                        provider_options=PROVIDER_OPTIONS,
+                    )
                     logger.info(f"Computed embeddings shape: {embeddings.shape}")
 
                     # Validation

packages/leann-backend-diskann/pyproject.toml

@@ -1,11 +1,11 @@
 [build-system]
-requires = ["scikit-build-core>=0.10", "pybind11>=2.12.0", "numpy"]
+requires = ["scikit-build-core>=0.10", "pybind11>=2.12.0", "numpy", "cmake>=3.30"]
 build-backend = "scikit_build_core.build"
 
 [project]
 name = "leann-backend-diskann"
-version = "0.3.2"
-dependencies = ["leann-core==0.3.2", "numpy", "protobuf>=3.19.0"]
+version = "0.3.4"
+dependencies = ["leann-core==0.3.4", "numpy", "protobuf>=3.19.0"]
 
 [tool.scikit-build]
 # Key: simplified CMake path

packages/leann-backend-hnsw/CMakeLists.txt

@@ -49,9 +49,28 @@ set(BUILD_TESTING OFF CACHE BOOL "" FORCE)
 set(FAISS_ENABLE_C_API OFF CACHE BOOL "" FORCE)
 set(FAISS_OPT_LEVEL "generic" CACHE STRING "" FORCE)
 
-# Disable additional SIMD versions to speed up compilation
+# Disable x86-specific SIMD optimizations (important for ARM64 compatibility)
 set(FAISS_ENABLE_AVX2 OFF CACHE BOOL "" FORCE)
 set(FAISS_ENABLE_AVX512 OFF CACHE BOOL "" FORCE)
+set(FAISS_ENABLE_SSE4_1 OFF CACHE BOOL "" FORCE)
+
+# ARM64-specific configuration
+if(CMAKE_SYSTEM_PROCESSOR MATCHES "aarch64|arm64")
+    message(STATUS "Configuring Faiss for ARM64 architecture")
+
+    if(CMAKE_SYSTEM_NAME STREQUAL "Linux")
+        # Use SVE optimization level for ARM64 Linux (as seen in Faiss conda build)
+        set(FAISS_OPT_LEVEL "sve" CACHE STRING "" FORCE)
+        message(STATUS "Setting FAISS_OPT_LEVEL to 'sve' for ARM64 Linux")
+    else()
+        # Use generic optimization for other ARM64 platforms (like macOS)
+        set(FAISS_OPT_LEVEL "generic" CACHE STRING "" FORCE)
+        message(STATUS "Setting FAISS_OPT_LEVEL to 'generic' for ARM64 ${CMAKE_SYSTEM_NAME}")
+    endif()
+
+    # ARM64 compatibility: Faiss submodule has been modified to fix x86 header inclusion
+    message(STATUS "Using ARM64-compatible Faiss submodule")
+endif()
 
 # Additional optimization options from INSTALL.md
 set(CMAKE_BUILD_TYPE "Release" CACHE STRING "" FORCE)

packages/leann-backend-hnsw/leann_backend_hnsw/convert_to_csr.py

@@ -5,6 +5,8 @@ import os
 import struct
 import sys
 import time
+from dataclasses import dataclass
+from typing import Any, Optional
 
 import numpy as np
 
@@ -237,6 +239,288 @@ def write_compact_format(
         f_out.write(storage_data)
 
 
+@dataclass
+class HNSWComponents:
+    original_hnsw_data: dict[str, Any]
+    assign_probas_np: np.ndarray
+    cum_nneighbor_per_level_np: np.ndarray
+    levels_np: np.ndarray
+    is_compact: bool
+    compact_level_ptr: Optional[np.ndarray] = None
+    compact_node_offsets_np: Optional[np.ndarray] = None
+    compact_neighbors_data: Optional[list[int]] = None
+    offsets_np: Optional[np.ndarray] = None
+    neighbors_np: Optional[np.ndarray] = None
+    storage_fourcc: int = NULL_INDEX_FOURCC
+    storage_data: bytes = b""
+
+
+def _read_hnsw_structure(f) -> HNSWComponents:
+    original_hnsw_data: dict[str, Any] = {}
+
+    hnsw_index_fourcc = read_struct(f, "<I")
+    if hnsw_index_fourcc not in EXPECTED_HNSW_FOURCCS:
+        raise ValueError(
+            f"Unexpected HNSW FourCC: {hnsw_index_fourcc:08x}. Expected one of {EXPECTED_HNSW_FOURCCS}."
+        )
+
+    original_hnsw_data["index_fourcc"] = hnsw_index_fourcc
+    original_hnsw_data["d"] = read_struct(f, "<i")
+    original_hnsw_data["ntotal"] = read_struct(f, "<q")
+    original_hnsw_data["dummy1"] = read_struct(f, "<q")
+    original_hnsw_data["dummy2"] = read_struct(f, "<q")
+    original_hnsw_data["is_trained"] = read_struct(f, "?")
+    original_hnsw_data["metric_type"] = read_struct(f, "<i")
+    original_hnsw_data["metric_arg"] = 0.0
+    if original_hnsw_data["metric_type"] > 1:
+        original_hnsw_data["metric_arg"] = read_struct(f, "<f")
+
+    assign_probas_np = read_numpy_vector(f, np.float64, "d")
+    cum_nneighbor_per_level_np = read_numpy_vector(f, np.int32, "i")
+    levels_np = read_numpy_vector(f, np.int32, "i")
+
+    ntotal = len(levels_np)
+    if ntotal != original_hnsw_data["ntotal"]:
+        original_hnsw_data["ntotal"] = ntotal
+
+    pos_before_compact = f.tell()
+    is_compact_flag = None
+    try:
+        is_compact_flag = read_struct(f, "<?")
+    except EOFError:
+        is_compact_flag = None
+
+    if is_compact_flag:
+        compact_level_ptr = read_numpy_vector(f, np.uint64, "Q")
+        compact_node_offsets_np = read_numpy_vector(f, np.uint64, "Q")
+
+        original_hnsw_data["entry_point"] = read_struct(f, "<i")
+        original_hnsw_data["max_level"] = read_struct(f, "<i")
+        original_hnsw_data["efConstruction"] = read_struct(f, "<i")
+        original_hnsw_data["efSearch"] = read_struct(f, "<i")
+        original_hnsw_data["dummy_upper_beam"] = read_struct(f, "<i")
+
+        storage_fourcc = read_struct(f, "<I")
+        compact_neighbors_data_np = read_numpy_vector(f, np.int32, "i")
+        compact_neighbors_data = compact_neighbors_data_np.tolist()
+        storage_data = f.read()
+
+        return HNSWComponents(
+            original_hnsw_data=original_hnsw_data,
+            assign_probas_np=assign_probas_np,
+            cum_nneighbor_per_level_np=cum_nneighbor_per_level_np,
+            levels_np=levels_np,
+            is_compact=True,
+            compact_level_ptr=compact_level_ptr,
+            compact_node_offsets_np=compact_node_offsets_np,
+            compact_neighbors_data=compact_neighbors_data,
+            storage_fourcc=storage_fourcc,
+            storage_data=storage_data,
+        )
+
+    # Non-compact case
+    f.seek(pos_before_compact)
+
+    pos_before_probe = f.tell()
+    try:
+        suspected_flag = read_struct(f, "<B")
+        if suspected_flag != 0x00:
+            f.seek(pos_before_probe)
+    except EOFError:
+        f.seek(pos_before_probe)
+
+    offsets_np = read_numpy_vector(f, np.uint64, "Q")
+    neighbors_np = read_numpy_vector(f, np.int32, "i")
+
+    original_hnsw_data["entry_point"] = read_struct(f, "<i")
+    original_hnsw_data["max_level"] = read_struct(f, "<i")
+    original_hnsw_data["efConstruction"] = read_struct(f, "<i")
+    original_hnsw_data["efSearch"] = read_struct(f, "<i")
+    original_hnsw_data["dummy_upper_beam"] = read_struct(f, "<i")
+
+    storage_fourcc = NULL_INDEX_FOURCC
+    storage_data = b""
+    try:
+        storage_fourcc = read_struct(f, "<I")
+        storage_data = f.read()
+    except EOFError:
+        storage_fourcc = NULL_INDEX_FOURCC
+
+    return HNSWComponents(
+        original_hnsw_data=original_hnsw_data,
+        assign_probas_np=assign_probas_np,
+        cum_nneighbor_per_level_np=cum_nneighbor_per_level_np,
+        levels_np=levels_np,
+        is_compact=False,
+        offsets_np=offsets_np,
+        neighbors_np=neighbors_np,
+        storage_fourcc=storage_fourcc,
+        storage_data=storage_data,
+    )
+
+
+def _read_hnsw_structure_from_file(path: str) -> HNSWComponents:
+    with open(path, "rb") as f:
+        return _read_hnsw_structure(f)
+
+
+def write_original_format(
+    f_out,
+    original_hnsw_data,
+    assign_probas_np,
+    cum_nneighbor_per_level_np,
+    levels_np,
+    offsets_np,
+    neighbors_np,
+    storage_fourcc,
+    storage_data,
+):
+    """Write non-compact HNSW data in original FAISS order."""
+
+    f_out.write(struct.pack("<I", original_hnsw_data["index_fourcc"]))
+    f_out.write(struct.pack("<i", original_hnsw_data["d"]))
+    f_out.write(struct.pack("<q", original_hnsw_data["ntotal"]))
+    f_out.write(struct.pack("<q", original_hnsw_data["dummy1"]))
+    f_out.write(struct.pack("<q", original_hnsw_data["dummy2"]))
+    f_out.write(struct.pack("<?", original_hnsw_data["is_trained"]))
+    f_out.write(struct.pack("<i", original_hnsw_data["metric_type"]))
+    if original_hnsw_data["metric_type"] > 1:
+        f_out.write(struct.pack("<f", original_hnsw_data["metric_arg"]))
+
+    write_numpy_vector(f_out, assign_probas_np, "d")
+    write_numpy_vector(f_out, cum_nneighbor_per_level_np, "i")
+    write_numpy_vector(f_out, levels_np, "i")
+
+    write_numpy_vector(f_out, offsets_np, "Q")
+    write_numpy_vector(f_out, neighbors_np, "i")
+
+    f_out.write(struct.pack("<i", original_hnsw_data["entry_point"]))
+    f_out.write(struct.pack("<i", original_hnsw_data["max_level"]))
+    f_out.write(struct.pack("<i", original_hnsw_data["efConstruction"]))
+    f_out.write(struct.pack("<i", original_hnsw_data["efSearch"]))
+    f_out.write(struct.pack("<i", original_hnsw_data["dummy_upper_beam"]))
+
+    f_out.write(struct.pack("<I", storage_fourcc))
+    if storage_fourcc != NULL_INDEX_FOURCC and storage_data:
+        f_out.write(storage_data)
+
+
+def prune_hnsw_embeddings(input_filename: str, output_filename: str) -> bool:
+    """Rewrite an HNSW index while dropping the embedded storage section."""
+
+    start_time = time.time()
+    try:
+        with open(input_filename, "rb") as f_in, open(output_filename, "wb") as f_out:
+            original_hnsw_data: dict[str, Any] = {}
+
+            hnsw_index_fourcc = read_struct(f_in, "<I")
+            if hnsw_index_fourcc not in EXPECTED_HNSW_FOURCCS:
+                print(
+                    f"Error: Expected HNSW Index FourCC ({list(EXPECTED_HNSW_FOURCCS)}), got {hnsw_index_fourcc:08x}.",
+                    file=sys.stderr,
+                )
+                return False
+
+            original_hnsw_data["index_fourcc"] = hnsw_index_fourcc
+            original_hnsw_data["d"] = read_struct(f_in, "<i")
+            original_hnsw_data["ntotal"] = read_struct(f_in, "<q")
+            original_hnsw_data["dummy1"] = read_struct(f_in, "<q")
+            original_hnsw_data["dummy2"] = read_struct(f_in, "<q")
+            original_hnsw_data["is_trained"] = read_struct(f_in, "?")
+            original_hnsw_data["metric_type"] = read_struct(f_in, "<i")
+            original_hnsw_data["metric_arg"] = 0.0
+            if original_hnsw_data["metric_type"] > 1:
+                original_hnsw_data["metric_arg"] = read_struct(f_in, "<f")
+
+            assign_probas_np = read_numpy_vector(f_in, np.float64, "d")
+            cum_nneighbor_per_level_np = read_numpy_vector(f_in, np.int32, "i")
+            levels_np = read_numpy_vector(f_in, np.int32, "i")
+
+            ntotal = len(levels_np)
+            if ntotal != original_hnsw_data["ntotal"]:
+                original_hnsw_data["ntotal"] = ntotal
+
+            pos_before_compact = f_in.tell()
+            is_compact_flag = None
+            try:
+                is_compact_flag = read_struct(f_in, "<?")
+            except EOFError:
+                is_compact_flag = None
+
+            if is_compact_flag:
+                compact_level_ptr = read_numpy_vector(f_in, np.uint64, "Q")
+                compact_node_offsets_np = read_numpy_vector(f_in, np.uint64, "Q")
+
+                original_hnsw_data["entry_point"] = read_struct(f_in, "<i")
+                original_hnsw_data["max_level"] = read_struct(f_in, "<i")
+                original_hnsw_data["efConstruction"] = read_struct(f_in, "<i")
+                original_hnsw_data["efSearch"] = read_struct(f_in, "<i")
+                original_hnsw_data["dummy_upper_beam"] = read_struct(f_in, "<i")
+
+                _storage_fourcc = read_struct(f_in, "<I")
+                compact_neighbors_data_np = read_numpy_vector(f_in, np.int32, "i")
+                compact_neighbors_data = compact_neighbors_data_np.tolist()
+                _storage_data = f_in.read()
+
+                write_compact_format(
+                    f_out,
+                    original_hnsw_data,
+                    assign_probas_np,
+                    cum_nneighbor_per_level_np,
+                    levels_np,
+                    compact_level_ptr,
+                    compact_node_offsets_np,
+                    compact_neighbors_data,
+                    NULL_INDEX_FOURCC,
+                    b"",
+                )
+            else:
+                f_in.seek(pos_before_compact)
+
+                pos_before_probe = f_in.tell()
+                try:
+                    suspected_flag = read_struct(f_in, "<B")
+                    if suspected_flag != 0x00:
+                        f_in.seek(pos_before_probe)
+                except EOFError:
+                    f_in.seek(pos_before_probe)
+
+                offsets_np = read_numpy_vector(f_in, np.uint64, "Q")
+                neighbors_np = read_numpy_vector(f_in, np.int32, "i")
+
+                original_hnsw_data["entry_point"] = read_struct(f_in, "<i")
+                original_hnsw_data["max_level"] = read_struct(f_in, "<i")
+                original_hnsw_data["efConstruction"] = read_struct(f_in, "<i")
+                original_hnsw_data["efSearch"] = read_struct(f_in, "<i")
+                original_hnsw_data["dummy_upper_beam"] = read_struct(f_in, "<i")
+
+                _storage_fourcc = None
+                _storage_data = b""
+                try:
+                    _storage_fourcc = read_struct(f_in, "<I")
+                    _storage_data = f_in.read()
+                except EOFError:
+                    _storage_fourcc = NULL_INDEX_FOURCC
+
+                write_original_format(
+                    f_out,
+                    original_hnsw_data,
+                    assign_probas_np,
+                    cum_nneighbor_per_level_np,
+                    levels_np,
+                    offsets_np,
+                    neighbors_np,
+                    NULL_INDEX_FOURCC,
+                    b"",
+                )
+
+        print(f"[{time.time() - start_time:.2f}s] Pruned embeddings from {input_filename}")
+        return True
+    except Exception as exc:
+        print(f"Failed to prune embeddings: {exc}", file=sys.stderr)
+        return False
+
+
 # --- Main Conversion Logic ---
 
 
@@ -700,6 +984,29 @@ def convert_hnsw_graph_to_csr(input_filename, output_filename, prune_embeddings=
             pass
 
 
+def prune_hnsw_embeddings_inplace(index_filename: str) -> bool:
+    """Convenience wrapper to prune embeddings in-place."""
+
+    temp_path = f"{index_filename}.prune.tmp"
+    success = prune_hnsw_embeddings(index_filename, temp_path)
+    if success:
+        try:
+            os.replace(temp_path, index_filename)
+        except Exception as exc:  # pragma: no cover - defensive
+            logger.error(f"Failed to replace original index with pruned version: {exc}")
+            try:
+                os.remove(temp_path)
+            except OSError:
+                pass
+            return False
+    else:
+        try:
+            os.remove(temp_path)
+        except OSError:
+            pass
+    return success
+
+
 # --- Script Execution ---
 if __name__ == "__main__":
     parser = argparse.ArgumentParser(

packages/leann-backend-hnsw/leann_backend_hnsw/hnsw_backend.py

@@ -14,7 +14,7 @@ from leann.interface import (
 from leann.registry import register_backend
 from leann.searcher_base import BaseSearcher
 
-from .convert_to_csr import convert_hnsw_graph_to_csr
+from .convert_to_csr import convert_hnsw_graph_to_csr, prune_hnsw_embeddings_inplace
 
 logger = logging.getLogger(__name__)
 
@@ -90,8 +90,19 @@ class HNSWBuilder(LeannBackendBuilderInterface):
         index_file = index_dir / f"{index_prefix}.index"
         faiss.write_index(index, str(index_file))
 
+        # Persist ID map so searcher can map FAISS integer labels back to passage IDs
+        try:
+            idmap_file = index_dir / f"{index_prefix}.ids.txt"
+            with open(idmap_file, "w", encoding="utf-8") as f:
+                for id_str in ids:
+                    f.write(str(id_str) + "\n")
+        except Exception as e:
+            logger.warning(f"Failed to write ID map: {e}")
+
         if self.is_compact:
             self._convert_to_csr(index_file)
+        elif self.is_recompute:
+            prune_hnsw_embeddings_inplace(str(index_file))
 
     def _convert_to_csr(self, index_file: Path):
         """Convert built index to CSR format"""
@@ -133,10 +144,10 @@ class HNSWSearcher(BaseSearcher):
         if metric_enum is None:
             raise ValueError(f"Unsupported distance_metric '{self.distance_metric}'.")
 
-        self.is_compact, self.is_pruned = (
-            self.meta.get("is_compact", True),
-            self.meta.get("is_pruned", True),
-        )
+        backend_meta_kwargs = self.meta.get("backend_kwargs", {})
+        self.is_compact = self.meta.get("is_compact", backend_meta_kwargs.get("is_compact", True))
+        default_pruned = backend_meta_kwargs.get("is_recompute", self.is_compact)
+        self.is_pruned = bool(self.meta.get("is_pruned", default_pruned))
 
         index_file = self.index_dir / f"{self.index_path.stem}.index"
         if not index_file.exists():
@@ -150,6 +161,16 @@ class HNSWSearcher(BaseSearcher):
 
         self._index = faiss.read_index(str(index_file), faiss.IO_FLAG_MMAP, hnsw_config)
 
+        # Load ID map if available
+        self._id_map: list[str] = []
+        try:
+            idmap_file = self.index_dir / f"{self.index_path.stem}.ids.txt"
+            if idmap_file.exists():
+                with open(idmap_file, encoding="utf-8") as f:
+                    self._id_map = [line.rstrip("\n") for line in f]
+        except Exception as e:
+            logger.warning(f"Failed to load ID map: {e}")
+
     def search(
         self,
         query: np.ndarray,
@@ -248,6 +269,19 @@ class HNSWSearcher(BaseSearcher):
         )
         search_time = time.time() - search_time
         logger.info(f"  Search time in HNSWSearcher.search() backend: {search_time} seconds")
-        string_labels = [[str(int_label) for int_label in batch_labels] for batch_labels in labels]
+        if self._id_map:
+
+            def map_label(x: int) -> str:
+                if 0 <= x < len(self._id_map):
+                    return self._id_map[x]
+                return str(x)
+
+            string_labels = [
+                [map_label(int(label)) for label in batch_labels] for batch_labels in labels
+            ]
+        else:
+            string_labels = [
+                [str(int_label) for int_label in batch_labels] for batch_labels in labels
+            ]
 
         return {"labels": string_labels, "distances": distances}

packages/leann-backend-hnsw/leann_backend_hnsw/hnsw_embedding_server.py

@@ -10,7 +10,7 @@ import sys
 import threading
 import time
 from pathlib import Path
-from typing import Optional
+from typing import Any, Optional
 
 import msgpack
 import numpy as np
@@ -24,14 +24,36 @@ logger = logging.getLogger(__name__)
 log_level = getattr(logging, LOG_LEVEL, logging.WARNING)
 logger.setLevel(log_level)
 
-# Ensure we have a handler if none exists
+# Ensure we have handlers if none exist
 if not logger.handlers:
-    handler = logging.StreamHandler()
+    stream_handler = logging.StreamHandler()
     formatter = logging.Formatter("%(asctime)s - %(levelname)s - %(message)s")
-    handler.setFormatter(formatter)
-    logger.addHandler(handler)
+    stream_handler.setFormatter(formatter)
+    logger.addHandler(stream_handler)
+
+log_path = os.getenv("LEANN_HNSW_LOG_PATH")
+if log_path:
+    try:
+        file_handler = logging.FileHandler(log_path, mode="a", encoding="utf-8")
+        file_formatter = logging.Formatter(
+            "%(asctime)s - %(levelname)s - [pid=%(process)d] %(message)s"
+        )
+        file_handler.setFormatter(file_formatter)
+        logger.addHandler(file_handler)
+    except Exception as exc:  # pragma: no cover - best effort logging
+        logger.warning(f"Failed to attach file handler for log path {log_path}: {exc}")
+
 logger.propagate = False
 
+_RAW_PROVIDER_OPTIONS = os.getenv("LEANN_EMBEDDING_OPTIONS")
+try:
+    PROVIDER_OPTIONS: dict[str, Any] = (
+        json.loads(_RAW_PROVIDER_OPTIONS) if _RAW_PROVIDER_OPTIONS else {}
+    )
+except json.JSONDecodeError:
+    logger.warning("Failed to parse LEANN_EMBEDDING_OPTIONS; ignoring provider options")
+    PROVIDER_OPTIONS = {}
+
 
 def create_hnsw_embedding_server(
     passages_file: Optional[str] = None,
@@ -92,6 +114,35 @@ def create_hnsw_embedding_server(
         embedding_dim = 0
     logger.info(f"Loaded PassageManager with {len(passages)} passages from metadata")
 
+    # Attempt to load ID map (maps FAISS integer labels -> passage IDs)
+    id_map: list[str] = []
+    try:
+        meta_path = Path(passages_file)
+        base = meta_path.name
+        if base.endswith(".meta.json"):
+            base = base[: -len(".meta.json")]  # e.g., laion_index.leann
+        if base.endswith(".leann"):
+            base = base[: -len(".leann")]  # e.g., laion_index
+        idmap_file = meta_path.parent / f"{base}.ids.txt"
+        if idmap_file.exists():
+            with open(idmap_file, encoding="utf-8") as f:
+                id_map = [line.rstrip("\n") for line in f]
+            logger.info(f"Loaded ID map with {len(id_map)} entries from {idmap_file}")
+        else:
+            logger.warning(f"ID map file not found at {idmap_file}; will use raw labels")
+    except Exception as e:
+        logger.warning(f"Failed to load ID map: {e}")
+
+    def _map_node_id(nid) -> str:
+        try:
+            if id_map is not None and len(id_map) > 0 and isinstance(nid, (int, np.integer)):
+                idx = int(nid)
+                if 0 <= idx < len(id_map):
+                    return id_map[idx]
+        except Exception:
+            pass
+        return str(nid)
+
     # (legacy ZMQ thread removed; using shutdown-capable server only)
 
     def zmq_server_thread_with_shutdown(shutdown_event):
@@ -138,7 +189,12 @@ def create_hnsw_embedding_server(
                     ):
                         last_request_type = "text"
                         last_request_length = len(request)
-                        embeddings = compute_embeddings(request, model_name, mode=embedding_mode)
+                        embeddings = compute_embeddings(
+                            request,
+                            model_name,
+                            mode=embedding_mode,
+                            provider_options=PROVIDER_OPTIONS,
+                        )
                         rep_socket.send(msgpack.packb(embeddings.tolist()))
                         e2e_end = time.time()
                         logger.info(f"⏱️  Text embedding E2E time: {e2e_end - e2e_start:.6f}s")
@@ -168,13 +224,14 @@ def create_hnsw_embedding_server(
                         found_indices: list[int] = []
                         for idx, nid in enumerate(node_ids):
                             try:
-                                passage_data = passages.get_passage(str(nid))
+                                passage_id = _map_node_id(nid)
+                                passage_data = passages.get_passage(passage_id)
                                 txt = passage_data.get("text", "")
                                 if isinstance(txt, str) and len(txt) > 0:
                                     texts.append(txt)
                                     found_indices.append(idx)
                                 else:
-                                    logger.error(f"Empty text for passage ID {nid}")
+                                    logger.error(f"Empty text for passage ID {passage_id}")
                             except KeyError:
                                 logger.error(f"Passage ID {nid} not found")
                             except Exception as e:
@@ -187,7 +244,10 @@ def create_hnsw_embedding_server(
                         if texts:
                             try:
                                 embeddings = compute_embeddings(
-                                    texts, model_name, mode=embedding_mode
+                                    texts,
+                                    model_name,
+                                    mode=embedding_mode,
+                                    provider_options=PROVIDER_OPTIONS,
                                 )
                                 logger.info(
                                     f"Computed embeddings for {len(texts)} texts, shape: {embeddings.shape}"
@@ -238,13 +298,14 @@ def create_hnsw_embedding_server(
                     found_indices: list[int] = []
                     for idx, nid in enumerate(node_ids):
                         try:
-                            passage_data = passages.get_passage(str(nid))
+                            passage_id = _map_node_id(nid)
+                            passage_data = passages.get_passage(passage_id)
                             txt = passage_data.get("text", "")
                             if isinstance(txt, str) and len(txt) > 0:
                                 texts.append(txt)
                                 found_indices.append(idx)
                             else:
-                                logger.error(f"Empty text for passage ID {nid}")
+                                logger.error(f"Empty text for passage ID {passage_id}")
                         except KeyError:
                             logger.error(f"Passage with ID {nid} not found")
                         except Exception as e:
@@ -252,7 +313,12 @@ def create_hnsw_embedding_server(
 
                     if texts:
                         try:
-                            embeddings = compute_embeddings(texts, model_name, mode=embedding_mode)
+                            embeddings = compute_embeddings(
+                                texts,
+                                model_name,
+                                mode=embedding_mode,
+                                provider_options=PROVIDER_OPTIONS,
+                            )
                             logger.info(
                                 f"Computed embeddings for {len(texts)} texts, shape: {embeddings.shape}"
                             )

packages/leann-backend-hnsw/pyproject.toml

@@ -6,10 +6,10 @@ build-backend = "scikit_build_core.build"
 
 [project]
 name = "leann-backend-hnsw"
-version = "0.3.2"
+version = "0.3.4"
 description = "Custom-built HNSW (Faiss) backend for the Leann toolkit."
 dependencies = [
-    "leann-core==0.3.2",
+    "leann-core==0.3.4",
     "numpy",
     "pyzmq>=23.0.0",
     "msgpack>=1.0.0",

packages/leann-core/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "leann-core"
-version = "0.3.2"
+version = "0.3.4"
 description = "Core API and plugin system for LEANN"
 readme = "README.md"
 requires-python = ">=3.9"
@@ -18,14 +18,16 @@ dependencies = [
     "pyzmq>=23.0.0",
     "msgpack>=1.0.0",
     "torch>=2.0.0",
-    "sentence-transformers>=2.2.0",
+    "sentence-transformers>=3.0.0",
     "llama-index-core>=0.12.0",
     "llama-index-readers-file>=0.4.0",  # Essential for document reading
     "llama-index-embeddings-huggingface>=0.5.5",  # For embeddings
     "python-dotenv>=1.0.0",
     "openai>=1.0.0",
     "huggingface-hub>=0.20.0",
-    "transformers>=4.30.0",
+    # Keep transformers below 4.46: 4.46.0 adds Python 3.10-only return type syntax and
+    # breaks Python 3.9 environments.
+    "transformers>=4.30.0,<4.46",
     "requests>=2.25.0",
     "accelerate>=0.20.0",
     "PyPDF2>=3.0.0",
@@ -40,7 +42,7 @@ dependencies = [
 [project.optional-dependencies]
 colab = [
     "torch>=2.0.0,<3.0.0",  # Limit torch version to avoid conflicts
-    "transformers>=4.30.0,<5.0.0",  # Limit transformers version
+    "transformers>=4.30.0,<4.46",  # 4.46.0 switches to PEP 604 typing (int | None), breaks Py3.9
     "accelerate>=0.20.0,<1.0.0",  # Limit accelerate version
 ]
 

packages/leann-core/src/leann/api.py

@@ -5,7 +5,10 @@ with the correct, original embedding logic from the user's reference code.
 
 import json
 import logging
+import os
 import pickle
+import re
+import subprocess
 import time
 import warnings
 from dataclasses import dataclass, field
@@ -13,10 +16,13 @@ from pathlib import Path
 from typing import Any, Literal, Optional, Union
 
 import numpy as np
+from leann_backend_hnsw.convert_to_csr import prune_hnsw_embeddings_inplace
 
+from leann.interactive_utils import create_api_session
 from leann.interface import LeannBackendSearcherInterface
 
 from .chat import get_llm
+from .embedding_server_manager import EmbeddingServerManager
 from .interface import LeannBackendFactoryInterface
 from .metadata_filter import MetadataFilterEngine
 from .registry import BACKEND_REGISTRY
@@ -36,6 +42,7 @@ def compute_embeddings(
     use_server: bool = True,
     port: Optional[int] = None,
     is_build=False,
+    provider_options: Optional[dict[str, Any]] = None,
 ) -> np.ndarray:
     """
     Computes embeddings using different backends.
@@ -69,6 +76,7 @@ def compute_embeddings(
             model_name,
             mode=mode,
             is_build=is_build,
+            provider_options=provider_options,
         )
 
 
@@ -275,6 +283,7 @@ class LeannBuilder:
         embedding_model: str = "facebook/contriever",
         dimensions: Optional[int] = None,
         embedding_mode: str = "sentence-transformers",
+        embedding_options: Optional[dict[str, Any]] = None,
         **backend_kwargs,
     ):
         self.backend_name = backend_name
@@ -297,6 +306,7 @@ class LeannBuilder:
         self.embedding_model = embedding_model
         self.dimensions = dimensions
         self.embedding_mode = embedding_mode
+        self.embedding_options = embedding_options or {}
 
         # Check if we need to use cosine distance for normalized embeddings
         normalized_embeddings_models = {
@@ -404,6 +414,7 @@ class LeannBuilder:
                     self.embedding_model,
                     self.embedding_mode,
                     use_server=False,
+                    provider_options=self.embedding_options,
                 )[0]
             )
         path = Path(index_path)
@@ -443,8 +454,20 @@ class LeannBuilder:
             self.embedding_mode,
             use_server=False,
             is_build=True,
+            provider_options=self.embedding_options,
         )
         string_ids = [chunk["id"] for chunk in self.chunks]
+        # Persist ID map alongside index so backends that return integer labels can remap to passage IDs
+        try:
+            idmap_file = (
+                index_dir
+                / f"{index_name[: -len('.leann')] if index_name.endswith('.leann') else index_name}.ids.txt"
+            )
+            with open(idmap_file, "w", encoding="utf-8") as f:
+                for sid in string_ids:
+                    f.write(str(sid) + "\n")
+        except Exception:
+            pass
         current_backend_kwargs = {**self.backend_kwargs, "dimensions": self.dimensions}
         builder_instance = self.backend_factory.builder(**current_backend_kwargs)
         builder_instance.build(embeddings, string_ids, index_path, **current_backend_kwargs)
@@ -469,14 +492,15 @@ class LeannBuilder:
             ],
         }
 
+        if self.embedding_options:
+            meta_data["embedding_options"] = self.embedding_options
+
         # Add storage status flags for HNSW backend
         if self.backend_name == "hnsw":
             is_compact = self.backend_kwargs.get("is_compact", True)
             is_recompute = self.backend_kwargs.get("is_recompute", True)
             meta_data["is_compact"] = is_compact
-            meta_data["is_pruned"] = (
-                is_compact and is_recompute
-            )  # Pruned only if compact and recompute
+            meta_data["is_pruned"] = bool(is_recompute)
         with open(leann_meta_path, "w", encoding="utf-8") as f:
             json.dump(meta_data, f, indent=2)
 
@@ -563,6 +587,17 @@ class LeannBuilder:
 
         # Build the vector index using precomputed embeddings
         string_ids = [str(id_val) for id_val in ids]
+        # Persist ID map (order == embeddings order)
+        try:
+            idmap_file = (
+                index_dir
+                / f"{index_name[: -len('.leann')] if index_name.endswith('.leann') else index_name}.ids.txt"
+            )
+            with open(idmap_file, "w", encoding="utf-8") as f:
+                for sid in string_ids:
+                    f.write(str(sid) + "\n")
+        except Exception:
+            pass
         current_backend_kwargs = {**self.backend_kwargs, "dimensions": self.dimensions}
         builder_instance = self.backend_factory.builder(**current_backend_kwargs)
         builder_instance.build(embeddings, string_ids, index_path)
@@ -591,18 +626,242 @@ class LeannBuilder:
             "embeddings_source": str(embeddings_file),
         }
 
+        if self.embedding_options:
+            meta_data["embedding_options"] = self.embedding_options
+
         # Add storage status flags for HNSW backend
         if self.backend_name == "hnsw":
             is_compact = self.backend_kwargs.get("is_compact", True)
             is_recompute = self.backend_kwargs.get("is_recompute", True)
             meta_data["is_compact"] = is_compact
-            meta_data["is_pruned"] = is_compact and is_recompute
+            meta_data["is_pruned"] = bool(is_recompute)
 
         with open(leann_meta_path, "w", encoding="utf-8") as f:
             json.dump(meta_data, f, indent=2)
 
         logger.info(f"Index built successfully from precomputed embeddings: {index_path}")
 
+    def update_index(self, index_path: str):
+        """Append new passages and vectors to an existing HNSW index."""
+        if not self.chunks:
+            raise ValueError("No new chunks provided for update.")
+
+        path = Path(index_path)
+        index_dir = path.parent
+        index_name = path.name
+        index_prefix = path.stem
+
+        meta_path = index_dir / f"{index_name}.meta.json"
+        passages_file = index_dir / f"{index_name}.passages.jsonl"
+        offset_file = index_dir / f"{index_name}.passages.idx"
+        index_file = index_dir / f"{index_prefix}.index"
+
+        if not meta_path.exists() or not passages_file.exists() or not offset_file.exists():
+            raise FileNotFoundError("Index metadata or passage files are missing; cannot update.")
+        if not index_file.exists():
+            raise FileNotFoundError(f"HNSW index file not found: {index_file}")
+
+        with open(meta_path, encoding="utf-8") as f:
+            meta = json.load(f)
+        backend_name = meta.get("backend_name")
+        if backend_name != self.backend_name:
+            raise ValueError(
+                f"Index was built with backend '{backend_name}', cannot update with '{self.backend_name}'."
+            )
+
+        meta_backend_kwargs = meta.get("backend_kwargs", {})
+        index_is_compact = meta.get("is_compact", meta_backend_kwargs.get("is_compact", True))
+        if index_is_compact:
+            raise ValueError(
+                "Compact HNSW indices do not support in-place updates. Rebuild required."
+            )
+
+        distance_metric = meta_backend_kwargs.get(
+            "distance_metric", self.backend_kwargs.get("distance_metric", "mips")
+        ).lower()
+        needs_recompute = bool(
+            meta.get("is_pruned")
+            or meta_backend_kwargs.get("is_recompute")
+            or self.backend_kwargs.get("is_recompute")
+        )
+
+        with open(offset_file, "rb") as f:
+            offset_map: dict[str, int] = pickle.load(f)
+        existing_ids = set(offset_map.keys())
+
+        valid_chunks: list[dict[str, Any]] = []
+        for chunk in self.chunks:
+            text = chunk.get("text", "")
+            if not isinstance(text, str) or not text.strip():
+                continue
+            metadata = chunk.setdefault("metadata", {})
+            passage_id = chunk.get("id") or metadata.get("id")
+            if passage_id and passage_id in existing_ids:
+                raise ValueError(f"Passage ID '{passage_id}' already exists in the index.")
+            valid_chunks.append(chunk)
+
+        if not valid_chunks:
+            raise ValueError("No valid chunks to append.")
+
+        texts_to_embed = [chunk["text"] for chunk in valid_chunks]
+        embeddings = compute_embeddings(
+            texts_to_embed,
+            self.embedding_model,
+            self.embedding_mode,
+            use_server=False,
+            is_build=True,
+            provider_options=self.embedding_options,
+        )
+
+        embedding_dim = embeddings.shape[1]
+        expected_dim = meta.get("dimensions")
+        if expected_dim is not None and expected_dim != embedding_dim:
+            raise ValueError(
+                f"Dimension mismatch during update: existing index uses {expected_dim}, got {embedding_dim}."
+            )
+
+        from leann_backend_hnsw import faiss  # type: ignore
+
+        embeddings = np.ascontiguousarray(embeddings, dtype=np.float32)
+        if distance_metric == "cosine":
+            norms = np.linalg.norm(embeddings, axis=1, keepdims=True)
+            norms[norms == 0] = 1
+            embeddings = embeddings / norms
+
+        index = faiss.read_index(str(index_file))
+        if hasattr(index, "is_recompute"):
+            index.is_recompute = needs_recompute
+            print(f"index.is_recompute: {index.is_recompute}")
+        if getattr(index, "storage", None) is None:
+            if index.metric_type == faiss.METRIC_INNER_PRODUCT:
+                storage_index = faiss.IndexFlatIP(index.d)
+            else:
+                storage_index = faiss.IndexFlatL2(index.d)
+            index.storage = storage_index
+            index.own_fields = True
+            # Faiss expects storage.ntotal to reflect the existing graph's
+            # population (even if the vectors themselves were pruned from disk
+            # for recompute mode).  When we attach a fresh IndexFlat here its
+            # ntotal starts at zero, which later causes IndexHNSW::add to
+            # believe new "preset" levels were provided and trips the
+            # `n0 + n == levels.size()` assertion.  Seed the temporary storage
+            # with the current ntotal so Faiss maintains the proper offset for
+            # incoming vectors.
+            try:
+                storage_index.ntotal = index.ntotal
+            except AttributeError:
+                # Older Faiss builds may not expose ntotal as a writable
+                # attribute; in that case we fall back to the default behaviour.
+                pass
+        if index.d != embedding_dim:
+            raise ValueError(
+                f"Existing index dimension ({index.d}) does not match new embeddings ({embedding_dim})."
+            )
+
+        passage_meta_mode = meta.get("embedding_mode", self.embedding_mode)
+        passage_provider_options = meta.get("embedding_options", self.embedding_options)
+
+        base_id = index.ntotal
+        for offset, chunk in enumerate(valid_chunks):
+            new_id = str(base_id + offset)
+            chunk.setdefault("metadata", {})["id"] = new_id
+            chunk["id"] = new_id
+
+        # Append passages/offsets before we attempt index.add so the ZMQ server
+        # can resolve newly assigned IDs during recompute. Keep rollback hooks
+        # so we can restore files if the update fails mid-way.
+        rollback_passages_size = passages_file.stat().st_size if passages_file.exists() else 0
+        offset_map_backup = offset_map.copy()
+
+        try:
+            with open(passages_file, "a", encoding="utf-8") as f:
+                for chunk in valid_chunks:
+                    offset = f.tell()
+                    json.dump(
+                        {
+                            "id": chunk["id"],
+                            "text": chunk["text"],
+                            "metadata": chunk.get("metadata", {}),
+                        },
+                        f,
+                        ensure_ascii=False,
+                    )
+                    f.write("\n")
+                    offset_map[chunk["id"]] = offset
+
+            with open(offset_file, "wb") as f:
+                pickle.dump(offset_map, f)
+
+            server_manager: Optional[EmbeddingServerManager] = None
+            server_started = False
+            requested_zmq_port = int(os.getenv("LEANN_UPDATE_ZMQ_PORT", "5557"))
+
+            try:
+                if needs_recompute:
+                    server_manager = EmbeddingServerManager(
+                        backend_module_name="leann_backend_hnsw.hnsw_embedding_server"
+                    )
+                    server_started, actual_port = server_manager.start_server(
+                        port=requested_zmq_port,
+                        model_name=self.embedding_model,
+                        embedding_mode=passage_meta_mode,
+                        passages_file=str(meta_path),
+                        distance_metric=distance_metric,
+                        provider_options=passage_provider_options,
+                    )
+                    if not server_started:
+                        raise RuntimeError(
+                            "Failed to start HNSW embedding server for recompute update."
+                        )
+                    if actual_port != requested_zmq_port:
+                        logger.warning(
+                            "Embedding server started on port %s instead of requested %s. "
+                            "Using reassigned port.",
+                            actual_port,
+                            requested_zmq_port,
+                        )
+                    try:
+                        index.hnsw.zmq_port = actual_port
+                    except AttributeError:
+                        pass
+
+                if needs_recompute:
+                    for i in range(embeddings.shape[0]):
+                        print(f"add {i} embeddings")
+                        index.add(1, faiss.swig_ptr(embeddings[i : i + 1]))
+                else:
+                    index.add(embeddings.shape[0], faiss.swig_ptr(embeddings))
+                faiss.write_index(index, str(index_file))
+            finally:
+                if server_started and server_manager is not None:
+                    server_manager.stop_server()
+
+        except Exception:
+            # Roll back appended passages/offset map to keep files consistent.
+            if passages_file.exists():
+                with open(passages_file, "rb+") as f:
+                    f.truncate(rollback_passages_size)
+            offset_map = offset_map_backup
+            with open(offset_file, "wb") as f:
+                pickle.dump(offset_map, f)
+            raise
+
+        meta["total_passages"] = len(offset_map)
+        with open(meta_path, "w", encoding="utf-8") as f:
+            json.dump(meta, f, indent=2)
+
+        logger.info(
+            "Appended %d passages to index '%s'. New total: %d",
+            len(valid_chunks),
+            index_path,
+            len(offset_map),
+        )
+
+        self.chunks.clear()
+
+        if needs_recompute:
+            prune_hnsw_embeddings_inplace(str(index_file))
+
 
 class LeannSearcher:
     def __init__(self, index_path: str, enable_warmup: bool = False, **backend_kwargs):
@@ -626,6 +885,7 @@ class LeannSearcher:
         self.embedding_model = self.meta_data["embedding_model"]
         # Support both old and new format
         self.embedding_mode = self.meta_data.get("embedding_mode", "sentence-transformers")
+        self.embedding_options = self.meta_data.get("embedding_options", {})
         # Delegate portability handling to PassageManager
         self.passage_manager = PassageManager(
             self.meta_data.get("passage_sources", []), metadata_file_path=self.meta_path_str
@@ -637,6 +897,8 @@ class LeannSearcher:
             raise ValueError(f"Backend '{backend_name}' not found.")
         final_kwargs = {**self.meta_data.get("backend_kwargs", {}), **backend_kwargs}
         final_kwargs["enable_warmup"] = enable_warmup
+        if self.embedding_options:
+            final_kwargs.setdefault("embedding_options", self.embedding_options)
         self.backend_impl: LeannBackendSearcherInterface = backend_factory.searcher(
             index_path, **final_kwargs
         )
@@ -653,6 +915,7 @@ class LeannSearcher:
         expected_zmq_port: int = 5557,
         metadata_filters: Optional[dict[str, dict[str, Union[str, int, float, bool, list]]]] = None,
         batch_size: int = 0,
+        use_grep: bool = False,
         **kwargs,
     ) -> list[SearchResult]:
         """
@@ -679,6 +942,10 @@ class LeannSearcher:
         Returns:
             List of SearchResult objects with text, metadata, and similarity scores
         """
+        # Handle grep search
+        if use_grep:
+            return self._grep_search(query, top_k)
+
         logger.info("🔍 LeannSearcher.search() called:")
         logger.info(f"  Query: '{query}'")
         logger.info(f"  Top_k: {top_k}")
@@ -795,9 +1062,96 @@ class LeannSearcher:
         logger.info(f"  {GREEN}✓ Final enriched results: {len(enriched_results)} passages{RESET}")
         return enriched_results
 
+    def _find_jsonl_file(self) -> Optional[str]:
+        """Find the .jsonl file containing raw passages for grep search"""
+        index_path = Path(self.meta_path_str).parent
+        potential_files = [
+            index_path / "documents.leann.passages.jsonl",
+            index_path.parent / "documents.leann.passages.jsonl",
+        ]
+
+        for file_path in potential_files:
+            if file_path.exists():
+                return str(file_path)
+        return None
+
+    def _grep_search(self, query: str, top_k: int = 5) -> list[SearchResult]:
+        """Perform grep-based search on raw passages"""
+        jsonl_file = self._find_jsonl_file()
+        if not jsonl_file:
+            raise FileNotFoundError("No .jsonl passages file found for grep search")
+
+        try:
+            cmd = ["grep", "-i", "-n", query, jsonl_file]
+            result = subprocess.run(cmd, capture_output=True, text=True, check=False)
+
+            if result.returncode == 1:
+                return []
+            elif result.returncode != 0:
+                raise RuntimeError(f"Grep failed: {result.stderr}")
+
+            matches = []
+            for line in result.stdout.strip().split("\n"):
+                if not line:
+                    continue
+                parts = line.split(":", 1)
+                if len(parts) != 2:
+                    continue
+
+                try:
+                    data = json.loads(parts[1])
+                    text = data.get("text", "")
+                    score = text.lower().count(query.lower())
+
+                    matches.append(
+                        SearchResult(
+                            id=data.get("id", parts[0]),
+                            text=text,
+                            metadata=data.get("metadata", {}),
+                            score=float(score),
+                        )
+                    )
+                except json.JSONDecodeError:
+                    continue
+
+            matches.sort(key=lambda x: x.score, reverse=True)
+            return matches[:top_k]
+
+        except FileNotFoundError:
+            raise RuntimeError(
+                "grep command not found. Please install grep or use semantic search."
+            )
+
+    def _python_regex_search(self, query: str, top_k: int = 5) -> list[SearchResult]:
+        """Fallback regex search"""
+        jsonl_file = self._find_jsonl_file()
+        if not jsonl_file:
+            raise FileNotFoundError("No .jsonl file found")
+
+        pattern = re.compile(re.escape(query), re.IGNORECASE)
+        matches = []
+
+        with open(jsonl_file, encoding="utf-8") as f:
+            for line_num, line in enumerate(f, 1):
+                if pattern.search(line):
+                    try:
+                        data = json.loads(line.strip())
+                        matches.append(
+                            SearchResult(
+                                id=data.get("id", str(line_num)),
+                                text=data.get("text", ""),
+                                metadata=data.get("metadata", {}),
+                                score=float(len(pattern.findall(data.get("text", "")))),
+                            )
+                        )
+                    except json.JSONDecodeError:
+                        continue
+
+        matches.sort(key=lambda x: x.score, reverse=True)
+        return matches[:top_k]
+
     def cleanup(self):
         """Explicitly cleanup embedding server resources.
-
         This method should be called after you're done using the searcher,
         especially in test environments or batch processing scenarios.
         """
@@ -853,6 +1207,7 @@ class LeannChat:
         expected_zmq_port: int = 5557,
         metadata_filters: Optional[dict[str, dict[str, Union[str, int, float, bool, list]]]] = None,
         batch_size: int = 0,
+        use_grep: bool = False,
         **search_kwargs,
     ):
         if llm_kwargs is None:
@@ -888,19 +1243,14 @@ class LeannChat:
         return ans
 
     def start_interactive(self):
-        print("\nLeann Chat started (type 'quit' to exit)")
-        while True:
-            try:
-                user_input = input("You: ").strip()
-                if user_input.lower() in ["quit", "exit"]:
-                    break
-                if not user_input:
-                    continue
+        """Start interactive chat session."""
+        session = create_api_session()
+
+        def handle_query(user_input: str):
             response = self.ask(user_input)
             print(f"Leann: {response}")
-            except (KeyboardInterrupt, EOFError):
-                print("\nGoodbye!")
-                break
+
+        session.run_interactive_loop(handle_query)
 
     def cleanup(self):
         """Explicitly cleanup embedding server resources.

packages/leann-core/src/leann/chat.py

@@ -12,6 +12,8 @@ from typing import Any, Optional
 
 import torch
 
+from .settings import resolve_ollama_host, resolve_openai_api_key, resolve_openai_base_url
+
 # Configure logging
 logging.basicConfig(level=logging.INFO)
 logger = logging.getLogger(__name__)
@@ -310,11 +312,12 @@ def search_hf_models(query: str, limit: int = 10) -> list[str]:
 
 
 def validate_model_and_suggest(
-    model_name: str, llm_type: str, host: str = "http://localhost:11434"
+    model_name: str, llm_type: str, host: Optional[str] = None
 ) -> Optional[str]:
     """Validate model name and provide suggestions if invalid"""
     if llm_type == "ollama":
-        available_models = check_ollama_models(host)
+        resolved_host = resolve_ollama_host(host)
+        available_models = check_ollama_models(resolved_host)
         if available_models and model_name not in available_models:
             error_msg = f"Model '{model_name}' not found in your local Ollama installation."
 
@@ -457,19 +460,19 @@ class LLMInterface(ABC):
 class OllamaChat(LLMInterface):
     """LLM interface for Ollama models."""
 
-    def __init__(self, model: str = "llama3:8b", host: str = "http://localhost:11434"):
+    def __init__(self, model: str = "llama3:8b", host: Optional[str] = None):
         self.model = model
-        self.host = host
-        logger.info(f"Initializing OllamaChat with model='{model}' and host='{host}'")
+        self.host = resolve_ollama_host(host)
+        logger.info(f"Initializing OllamaChat with model='{model}' and host='{self.host}'")
         try:
             import requests
 
             # Check if the Ollama server is responsive
-            if host:
-                requests.get(host)
+            if self.host:
+                requests.get(self.host)
 
             # Pre-check model availability with helpful suggestions
-            model_error = validate_model_and_suggest(model, "ollama", host)
+            model_error = validate_model_and_suggest(model, "ollama", self.host)
             if model_error:
                 raise ValueError(model_error)
 
@@ -478,9 +481,11 @@ class OllamaChat(LLMInterface):
                 "The 'requests' library is required for Ollama. Please install it with 'pip install requests'."
             )
         except requests.exceptions.ConnectionError:
-            logger.error(f"Could not connect to Ollama at {host}. Please ensure Ollama is running.")
+            logger.error(
+                f"Could not connect to Ollama at {self.host}. Please ensure Ollama is running."
+            )
             raise ConnectionError(
-                f"Could not connect to Ollama at {host}. Please ensure Ollama is running."
+                f"Could not connect to Ollama at {self.host}. Please ensure Ollama is running."
             )
 
     def ask(self, prompt: str, **kwargs) -> str:
@@ -522,8 +527,6 @@ class OllamaChat(LLMInterface):
         logger.debug(f"Sending request to Ollama: {payload}")
         try:
             logger.info("Sending request to Ollama and waiting for response...")
-            max_tokens = kwargs.get("max_tokens", 1000)
-            payload["options"]["max_tokens"] = max_tokens
             response = requests.post(full_url, data=json.dumps(payload))
             response.raise_for_status()
 
@@ -543,11 +546,30 @@ class OllamaChat(LLMInterface):
 
 
 class HFChat(LLMInterface):
-    """LLM interface for local Hugging Face Transformers models with proper chat templates."""
+    """LLM interface for local Hugging Face Transformers models with proper chat templates.
 
-    def __init__(self, model_name: str = "deepseek-ai/deepseek-llm-7b-chat"):
+    Args:
+        model_name (str): Name of the Hugging Face model to load.
+        trust_remote_code (bool): Whether to allow execution of code from the model repository.
+            Defaults to False for security. Only enable for trusted models as this can pose
+            a security risk if the model repository is compromised.
+    """
+
+    def __init__(
+        self, model_name: str = "deepseek-ai/deepseek-llm-7b-chat", trust_remote_code: bool = False
+    ):
         logger.info(f"Initializing HFChat with model='{model_name}'")
 
+        # Security warning when trust_remote_code is enabled
+        if trust_remote_code:
+            logger.warning(
+                "SECURITY WARNING: trust_remote_code=True allows execution of arbitrary code from the model repository. "
+                "Only enable this for models from trusted sources. This creates a potential security risk if the model "
+                "repository is compromised."
+            )
+
+        self.trust_remote_code = trust_remote_code
+
         # Pre-check model availability with helpful suggestions
         model_error = validate_model_and_suggest(model_name, "hf")
         if model_error:
@@ -585,14 +607,16 @@ class HFChat(LLMInterface):
 
             try:
                 logger.info(f"Loading tokenizer for {model_name}...")
-                self.tokenizer = AutoTokenizer.from_pretrained(model_name)
+                self.tokenizer = AutoTokenizer.from_pretrained(
+                    model_name, trust_remote_code=self.trust_remote_code
+                )
 
                 logger.info(f"Loading model {model_name}...")
                 self.model = AutoModelForCausalLM.from_pretrained(
                     model_name,
                     torch_dtype=torch.float16 if self.device != "cpu" else torch.float32,
                     device_map="auto" if self.device != "cpu" else None,
-                    trust_remote_code=True,
+                    trust_remote_code=self.trust_remote_code,
                 )
                 logger.info(f"Successfully loaded {model_name}")
             finally:
@@ -622,8 +646,8 @@ class HFChat(LLMInterface):
         is_qwen_model = "qwen" in self.model.config._name_or_path.lower()
 
         # For Qwen models, automatically add /no_think to the prompt
-        # if is_qwen_model and "/no_think" not in prompt and "/think" not in prompt:
-        #     prompt = prompt + " /no_think"
+        if is_qwen_model and "/no_think" not in prompt and "/think" not in prompt:
+            prompt = prompt + " /no_think"
 
         # Prepare chat template
         messages = [{"role": "user", "content": prompt}]
@@ -739,21 +763,31 @@ class GeminiChat(LLMInterface):
 class OpenAIChat(LLMInterface):
     """LLM interface for OpenAI models."""
 
-    def __init__(self, model: str = "gpt-4o", api_key: Optional[str] = None):
+    def __init__(
+        self,
+        model: str = "gpt-4o",
+        api_key: Optional[str] = None,
+        base_url: Optional[str] = None,
+    ):
         self.model = model
-        self.api_key = api_key or os.getenv("OPENAI_API_KEY")
+        self.base_url = resolve_openai_base_url(base_url)
+        self.api_key = resolve_openai_api_key(api_key)
 
         if not self.api_key:
             raise ValueError(
                 "OpenAI API key is required. Set OPENAI_API_KEY environment variable or pass api_key parameter."
             )
 
-        logger.info(f"Initializing OpenAI Chat with model='{model}'")
+        logger.info(
+            "Initializing OpenAI Chat with model='%s' and base_url='%s'",
+            model,
+            self.base_url,
+        )
 
         try:
             import openai
 
-            self.client = openai.OpenAI(api_key=self.api_key)
+            self.client = openai.OpenAI(api_key=self.api_key, base_url=self.base_url)
         except ImportError:
             raise ImportError(
                 "The 'openai' library is required for OpenAI models. Please install it with 'pip install openai'."
@@ -843,12 +877,19 @@ def get_llm(llm_config: Optional[dict[str, Any]] = None) -> LLMInterface:
     if llm_type == "ollama":
         return OllamaChat(
             model=model or "llama3:8b",
-            host=llm_config.get("host", "http://localhost:11434"),
+            host=llm_config.get("host"),
         )
     elif llm_type == "hf":
-        return HFChat(model_name=model or "deepseek-ai/deepseek-llm-7b-chat")
+        return HFChat(
+            model_name=model or "deepseek-ai/deepseek-llm-7b-chat",
+            trust_remote_code=llm_config.get("trust_remote_code", False),
+        )
     elif llm_type == "openai":
-        return OpenAIChat(model=model or "gpt-4o", api_key=llm_config.get("api_key"))
+        return OpenAIChat(
+            model=model or "gpt-4o",
+            api_key=llm_config.get("api_key"),
+            base_url=llm_config.get("base_url"),
+        )
     elif llm_type == "gemini":
         return GeminiChat(model=model or "gemini-2.5-flash", api_key=llm_config.get("api_key"))
     elif llm_type == "simulated":

packages/leann-core/src/leann/chunking_utils.py

@@ -1,6 +1,6 @@
 """
 Enhanced chunking utilities with AST-aware code chunking support.
-Provides unified interface for both traditional and AST-based text chunking.
+Packaged within leann-core so installed wheels can import it reliably.
 """
 
 import logging
@@ -22,30 +22,9 @@ CODE_EXTENSIONS = {
     ".jsx": "typescript",
 }
 
-# Default chunk parameters for different content types
-DEFAULT_CHUNK_PARAMS = {
-    "code": {
-        "max_chunk_size": 512,
-        "chunk_overlap": 64,
-    },
-    "text": {
-        "chunk_size": 256,
-        "chunk_overlap": 128,
-    },
-}
-
 
 def detect_code_files(documents, code_extensions=None) -> tuple[list, list]:
-    """
-    Separate documents into code files and regular text files.
-
-    Args:
-        documents: List of LlamaIndex Document objects
-        code_extensions: Dict mapping file extensions to languages (defaults to CODE_EXTENSIONS)
-
-    Returns:
-        Tuple of (code_documents, text_documents)
-    """
+    """Separate documents into code files and regular text files."""
     if code_extensions is None:
         code_extensions = CODE_EXTENSIONS
 
@@ -53,16 +32,10 @@ def detect_code_files(documents, code_extensions=None) -> tuple[list, list]:
     text_docs = []
 
     for doc in documents:
-        # Get file path from metadata
-        file_path = doc.metadata.get("file_path", "")
-        if not file_path:
-            # Fallback to file_name
-            file_path = doc.metadata.get("file_name", "")
-
+        file_path = doc.metadata.get("file_path", "") or doc.metadata.get("file_name", "")
         if file_path:
             file_ext = Path(file_path).suffix.lower()
             if file_ext in code_extensions:
-                # Add language info to metadata
                 doc.metadata["language"] = code_extensions[file_ext]
                 doc.metadata["is_code"] = True
                 code_docs.append(doc)
@@ -70,7 +43,6 @@ def detect_code_files(documents, code_extensions=None) -> tuple[list, list]:
                 doc.metadata["is_code"] = False
                 text_docs.append(doc)
         else:
-            # If no file path, treat as text
             doc.metadata["is_code"] = False
             text_docs.append(doc)
 
@@ -79,7 +51,7 @@ def detect_code_files(documents, code_extensions=None) -> tuple[list, list]:
 
 
 def get_language_from_extension(file_path: str) -> Optional[str]:
-    """Get the programming language from file extension."""
+    """Return language string from a filename/extension using CODE_EXTENSIONS."""
     ext = Path(file_path).suffix.lower()
     return CODE_EXTENSIONS.get(ext)
 
@@ -90,40 +62,26 @@ def create_ast_chunks(
     chunk_overlap: int = 64,
     metadata_template: str = "default",
 ) -> list[str]:
-    """
-    Create AST-aware chunks from code documents using astchunk.
+    """Create AST-aware chunks from code documents using astchunk.
 
-    Args:
-        documents: List of code documents
-        max_chunk_size: Maximum characters per chunk
-        chunk_overlap: Number of AST nodes to overlap between chunks
-        metadata_template: Template for chunk metadata
-
-    Returns:
-        List of text chunks with preserved code structure
+    Falls back to traditional chunking if astchunk is unavailable.
     """
     try:
-        from astchunk import ASTChunkBuilder
+        from astchunk import ASTChunkBuilder  # optional dependency
     except ImportError as e:
         logger.error(f"astchunk not available: {e}")
         logger.info("Falling back to traditional chunking for code files")
         return create_traditional_chunks(documents, max_chunk_size, chunk_overlap)
 
     all_chunks = []
-
     for doc in documents:
-        # Get language from metadata (set by detect_code_files)
         language = doc.metadata.get("language")
         if not language:
-            logger.warning(
-                "No language detected for document, falling back to traditional chunking"
-            )
-            traditional_chunks = create_traditional_chunks([doc], max_chunk_size, chunk_overlap)
-            all_chunks.extend(traditional_chunks)
+            logger.warning("No language detected; falling back to traditional chunking")
+            all_chunks.extend(create_traditional_chunks([doc], max_chunk_size, chunk_overlap))
             continue
 
         try:
-            # Configure astchunk
             configs = {
                 "max_chunk_size": max_chunk_size,
                 "language": language,
@@ -131,7 +89,6 @@ def create_ast_chunks(
                 "chunk_overlap": chunk_overlap if chunk_overlap > 0 else 0,
             }
 
-            # Add repository-level metadata if available
             repo_metadata = {
                 "file_path": doc.metadata.get("file_path", ""),
                 "file_name": doc.metadata.get("file_name", ""),
@@ -140,17 +97,13 @@ def create_ast_chunks(
             }
             configs["repo_level_metadata"] = repo_metadata
 
-            # Create chunk builder and process
             chunk_builder = ASTChunkBuilder(**configs)
             code_content = doc.get_content()
-
             if not code_content or not code_content.strip():
                 logger.warning("Empty code content, skipping")
                 continue
 
             chunks = chunk_builder.chunkify(code_content)
-
-            # Extract text content from chunks
             for chunk in chunks:
                 if hasattr(chunk, "text"):
                     chunk_text = chunk.text
@@ -159,7 +112,6 @@ def create_ast_chunks(
                 elif isinstance(chunk, str):
                     chunk_text = chunk
                 else:
-                    # Try to convert to string
                     chunk_text = str(chunk)
 
                 if chunk_text and chunk_text.strip():
@@ -168,12 +120,10 @@ def create_ast_chunks(
             logger.info(
                 f"Created {len(chunks)} AST chunks from {language} file: {doc.metadata.get('file_name', 'unknown')}"
             )
-
         except Exception as e:
             logger.warning(f"AST chunking failed for {language} file: {e}")
             logger.info("Falling back to traditional chunking")
-            traditional_chunks = create_traditional_chunks([doc], max_chunk_size, chunk_overlap)
-            all_chunks.extend(traditional_chunks)
+            all_chunks.extend(create_traditional_chunks([doc], max_chunk_size, chunk_overlap))
 
     return all_chunks
 
@@ -181,23 +131,10 @@ def create_ast_chunks(
 def create_traditional_chunks(
     documents, chunk_size: int = 256, chunk_overlap: int = 128
 ) -> list[str]:
-    """
-    Create traditional text chunks using LlamaIndex SentenceSplitter.
-
-    Args:
-        documents: List of documents to chunk
-        chunk_size: Size of each chunk in characters
-        chunk_overlap: Overlap between chunks
-
-    Returns:
-        List of text chunks
-    """
-    # Handle invalid chunk_size values
+    """Create traditional text chunks using LlamaIndex SentenceSplitter."""
     if chunk_size <= 0:
         logger.warning(f"Invalid chunk_size={chunk_size}, using default value of 256")
         chunk_size = 256
-
-    # Ensure chunk_overlap is not negative and not larger than chunk_size
     if chunk_overlap < 0:
         chunk_overlap = 0
     if chunk_overlap >= chunk_size:
@@ -215,12 +152,9 @@ def create_traditional_chunks(
         try:
             nodes = node_parser.get_nodes_from_documents([doc])
             if nodes:
-                chunk_texts = [node.get_content() for node in nodes]
-                all_texts.extend(chunk_texts)
-                logger.debug(f"Created {len(chunk_texts)} traditional chunks from document")
+                all_texts.extend(node.get_content() for node in nodes)
         except Exception as e:
             logger.error(f"Traditional chunking failed for document: {e}")
-            # As last resort, add the raw content
             content = doc.get_content()
             if content and content.strip():
                 all_texts.append(content.strip())
@@ -238,32 +172,13 @@ def create_text_chunks(
     code_file_extensions: Optional[list[str]] = None,
     ast_fallback_traditional: bool = True,
 ) -> list[str]:
-    """
-    Create text chunks from documents with optional AST support for code files.
-
-    Args:
-        documents: List of LlamaIndex Document objects
-        chunk_size: Size for traditional text chunks
-        chunk_overlap: Overlap for traditional text chunks
-        use_ast_chunking: Whether to use AST chunking for code files
-        ast_chunk_size: Size for AST chunks
-        ast_chunk_overlap: Overlap for AST chunks
-        code_file_extensions: Custom list of code file extensions
-        ast_fallback_traditional: Fall back to traditional chunking on AST errors
-
-    Returns:
-        List of text chunks
-    """
+    """Create text chunks from documents with optional AST support for code files."""
     if not documents:
         logger.warning("No documents provided for chunking")
         return []
 
-    # Create a local copy of supported extensions for this function call
     local_code_extensions = CODE_EXTENSIONS.copy()
-
-    # Update supported extensions if provided
     if code_file_extensions:
-        # Map extensions to languages (simplified mapping)
         ext_mapping = {
             ".py": "python",
             ".java": "java",
@@ -273,47 +188,32 @@ def create_text_chunks(
         }
         for ext in code_file_extensions:
             if ext.lower() not in local_code_extensions:
-                # Try to guess language from extension
                 if ext.lower() in ext_mapping:
                     local_code_extensions[ext.lower()] = ext_mapping[ext.lower()]
                 else:
                     logger.warning(f"Unsupported extension {ext}, will use traditional chunking")
 
     all_chunks = []
-
     if use_ast_chunking:
-        # Separate code and text documents using local extensions
         code_docs, text_docs = detect_code_files(documents, local_code_extensions)
-
-        # Process code files with AST chunking
         if code_docs:
-            logger.info(f"Processing {len(code_docs)} code files with AST chunking")
             try:
-                ast_chunks = create_ast_chunks(
+                all_chunks.extend(
+                    create_ast_chunks(
                         code_docs, max_chunk_size=ast_chunk_size, chunk_overlap=ast_chunk_overlap
                     )
-                all_chunks.extend(ast_chunks)
-                logger.info(f"Created {len(ast_chunks)} AST chunks from code files")
+                )
             except Exception as e:
                 logger.error(f"AST chunking failed: {e}")
                 if ast_fallback_traditional:
-                    logger.info("Falling back to traditional chunking for code files")
-                    traditional_code_chunks = create_traditional_chunks(
-                        code_docs, chunk_size, chunk_overlap
+                    all_chunks.extend(
+                        create_traditional_chunks(code_docs, chunk_size, chunk_overlap)
                     )
-                    all_chunks.extend(traditional_code_chunks)
                 else:
                     raise
-
-        # Process text files with traditional chunking
         if text_docs:
-            logger.info(f"Processing {len(text_docs)} text files with traditional chunking")
-            text_chunks = create_traditional_chunks(text_docs, chunk_size, chunk_overlap)
-            all_chunks.extend(text_chunks)
-            logger.info(f"Created {len(text_chunks)} traditional chunks from text files")
+            all_chunks.extend(create_traditional_chunks(text_docs, chunk_size, chunk_overlap))
     else:
-        # Use traditional chunking for all files
-        logger.info(f"Processing {len(documents)} documents with traditional chunking")
         all_chunks = create_traditional_chunks(documents, chunk_size, chunk_overlap)
 
     logger.info(f"Total chunks created: {len(all_chunks)}")

packages/leann-core/src/leann/cli.py

@@ -1,6 +1,5 @@
 import argparse
 import asyncio
-import sys
 from pathlib import Path
 from typing import Any, Optional, Union
 
@@ -9,7 +8,9 @@ from llama_index.core.node_parser import SentenceSplitter
 from tqdm import tqdm
 
 from .api import LeannBuilder, LeannChat, LeannSearcher
+from .interactive_utils import create_cli_session
 from .registry import register_project_directory
+from .settings import resolve_ollama_host, resolve_openai_api_key, resolve_openai_base_url
 
 
 def extract_pdf_text_with_pymupdf(file_path: str) -> str:
@@ -124,6 +125,24 @@ Examples:
             choices=["sentence-transformers", "openai", "mlx", "ollama"],
             help="Embedding backend mode (default: sentence-transformers)",
         )
+        build_parser.add_argument(
+            "--embedding-host",
+            type=str,
+            default=None,
+            help="Override Ollama-compatible embedding host",
+        )
+        build_parser.add_argument(
+            "--embedding-api-base",
+            type=str,
+            default=None,
+            help="Base URL for OpenAI-compatible embedding services",
+        )
+        build_parser.add_argument(
+            "--embedding-api-key",
+            type=str,
+            default=None,
+            help="API key for embedding service (defaults to OPENAI_API_KEY)",
+        )
         build_parser.add_argument(
             "--force", "-f", action="store_true", help="Force rebuild existing index"
         )
@@ -239,6 +258,11 @@ Examples:
         # Ask command
         ask_parser = subparsers.add_parser("ask", help="Ask questions")
         ask_parser.add_argument("index_name", help="Index name")
+        ask_parser.add_argument(
+            "query",
+            nargs="?",
+            help="Question to ask (omit for prompt or when using --interactive)",
+        )
         ask_parser.add_argument(
             "--llm",
             type=str,
@@ -249,7 +273,12 @@ Examples:
         ask_parser.add_argument(
             "--model", type=str, default="qwen3:8b", help="Model name (default: qwen3:8b)"
         )
-        ask_parser.add_argument("--host", type=str, default="http://localhost:11434")
+        ask_parser.add_argument(
+            "--host",
+            type=str,
+            default=None,
+            help="Override Ollama-compatible host (defaults to LEANN_OLLAMA_HOST/OLLAMA_HOST)",
+        )
         ask_parser.add_argument(
             "--interactive", "-i", action="store_true", help="Interactive chat mode"
         )
@@ -278,6 +307,18 @@ Examples:
             default=None,
             help="Thinking budget for reasoning models (low/medium/high). Supported by GPT-Oss:20b and other reasoning models.",
         )
+        ask_parser.add_argument(
+            "--api-base",
+            type=str,
+            default=None,
+            help="Base URL for OpenAI-compatible APIs (e.g., http://localhost:10000/v1)",
+        )
+        ask_parser.add_argument(
+            "--api-key",
+            type=str,
+            default=None,
+            help="API key for OpenAI-compatible APIs (defaults to OPENAI_API_KEY)",
+        )
 
         # List command
         subparsers.add_parser("list", help="List all indexes")
@@ -322,9 +363,17 @@ Examples:
 
         return basic_matches
 
-    def _should_exclude_file(self, relative_path: Path, gitignore_matches) -> bool:
-        """Check if a file should be excluded using gitignore parser."""
-        return gitignore_matches(str(relative_path))
+    def _should_exclude_file(self, file_path: Path, gitignore_matches) -> bool:
+        """Check if a file should be excluded using gitignore parser.
+
+        Always match against absolute, posix-style paths for consistency with
+        gitignore_parser expectations.
+        """
+        try:
+            absolute_path = file_path.resolve()
+        except Exception:
+            absolute_path = Path(str(file_path))
+        return gitignore_matches(absolute_path.as_posix())
 
     def _is_git_submodule(self, path: Path) -> bool:
         """Check if a path is a git submodule."""
@@ -396,7 +445,9 @@ Examples:
         print(f"   {current_path}")
         print("   " + "─" * 45)
 
-        current_indexes = self._discover_indexes_in_project(current_path)
+        current_indexes = self._discover_indexes_in_project(
+            current_path, exclude_dirs=other_projects
+        )
         if current_indexes:
             for idx in current_indexes:
                 total_indexes += 1
@@ -435,9 +486,14 @@ Examples:
             print("   leann build my-docs --docs ./documents")
         else:
             # Count only projects that have at least one discoverable index
-            projects_count = sum(
-                1 for p in valid_projects if len(self._discover_indexes_in_project(p)) > 0
-            )
+            projects_count = 0
+            for p in valid_projects:
+                if p == current_path:
+                    discovered = self._discover_indexes_in_project(p, exclude_dirs=other_projects)
+                else:
+                    discovered = self._discover_indexes_in_project(p)
+                if len(discovered) > 0:
+                    projects_count += 1
             print(f"📊 Total: {total_indexes} indexes across {projects_count} projects")
 
             if current_indexes_count > 0:
@@ -454,9 +510,22 @@ Examples:
                 print("\n💡 Create your first index:")
                 print("   leann build my-docs --docs ./documents")
 
-    def _discover_indexes_in_project(self, project_path: Path):
-        """Discover all indexes in a project directory (both CLI and apps formats)"""
+    def _discover_indexes_in_project(
+        self, project_path: Path, exclude_dirs: Optional[list[Path]] = None
+    ):
+        """Discover all indexes in a project directory (both CLI and apps formats)
+
+        exclude_dirs: when provided, skip any APP-format index files that are
+        located under these directories. This prevents duplicates when the
+        current project is a parent directory of other registered projects.
+        """
         indexes = []
+        exclude_dirs = exclude_dirs or []
+        # normalize to resolved paths once for comparison
+        try:
+            exclude_dirs_resolved = [p.resolve() for p in exclude_dirs]
+        except Exception:
+            exclude_dirs_resolved = exclude_dirs
 
         # 1. CLI format: .leann/indexes/index_name/
         cli_indexes_dir = project_path / ".leann" / "indexes"
@@ -495,6 +564,17 @@ Examples:
                         continue
                 except Exception:
                     pass
+                # Skip meta files that live under excluded directories
+                try:
+                    meta_parent_resolved = meta_file.parent.resolve()
+                    if any(
+                        meta_parent_resolved.is_relative_to(ex_dir)
+                        for ex_dir in exclude_dirs_resolved
+                    ):
+                        continue
+                except Exception:
+                    # best effort; if resolve or comparison fails, do not exclude
+                    pass
                 # Use the parent directory name as the app index display name
                 display_name = meta_file.parent.name
                 # Extract file base used to store files
@@ -1022,7 +1102,8 @@ Examples:
 
             # Try to use better PDF parsers first, but only if PDFs are requested
             documents = []
-            docs_path = Path(docs_dir)
+            # Use resolved absolute paths to avoid mismatches (symlinks, relative vs absolute)
+            docs_path = Path(docs_dir).resolve()
 
             # Check if we should process PDFs
             should_process_pdfs = custom_file_types is None or ".pdf" in custom_file_types
@@ -1031,10 +1112,15 @@ Examples:
                 for file_path in docs_path.rglob("*.pdf"):
                     # Check if file matches any exclude pattern
                     try:
+                        # Ensure both paths are resolved before computing relativity
+                        file_path_resolved = file_path.resolve()
+                        # Determine directory scope using the non-resolved path to avoid
+                        # misclassifying symlinked entries as outside the docs directory
                         relative_path = file_path.relative_to(docs_path)
                         if not include_hidden and _path_has_hidden_segment(relative_path):
                             continue
-                        if self._should_exclude_file(relative_path, gitignore_matches):
+                        # Use absolute path for gitignore matching
+                        if self._should_exclude_file(file_path_resolved, gitignore_matches):
                             continue
                     except ValueError:
                         # Skip files that can't be made relative to docs_path
@@ -1077,10 +1163,11 @@ Examples:
                 ) -> bool:
                     """Return True if file should be included (not excluded)"""
                     try:
-                        docs_path_obj = Path(docs_dir)
-                        file_path_obj = Path(file_path)
-                        relative_path = file_path_obj.relative_to(docs_path_obj)
-                        return not self._should_exclude_file(relative_path, gitignore_matches)
+                        docs_path_obj = Path(docs_dir).resolve()
+                        file_path_obj = Path(file_path).resolve()
+                        # Use absolute path for gitignore matching
+                        _ = file_path_obj.relative_to(docs_path_obj)  # validate scope
+                        return not self._should_exclude_file(file_path_obj, gitignore_matches)
                     except (ValueError, OSError):
                         return True  # Include files that can't be processed
 
@@ -1170,13 +1257,8 @@ Examples:
         if use_ast:
             print("🧠 Using AST-aware chunking for code files")
             try:
-                # Import enhanced chunking utilities
-                # Add apps directory to path to import chunking utilities
-                apps_dir = Path(__file__).parent.parent.parent.parent.parent / "apps"
-                if apps_dir.exists():
-                    sys.path.insert(0, str(apps_dir))
-
-                from chunking import create_text_chunks
+                # Import enhanced chunking utilities from packaged module
+                from .chunking_utils import create_text_chunks
 
                 # Use enhanced chunking with AST support
                 all_texts = create_text_chunks(
@@ -1191,7 +1273,9 @@ Examples:
                 )
 
             except ImportError as e:
-                print(f"⚠️  AST chunking not available ({e}), falling back to traditional chunking")
+                print(
+                    f"⚠️  AST chunking utilities not available in package ({e}), falling back to traditional chunking"
+                )
                 use_ast = False
 
         if not use_ast:
@@ -1283,10 +1367,20 @@ Examples:
 
         print(f"Building index '{index_name}' with {args.backend} backend...")
 
+        embedding_options: dict[str, Any] = {}
+        if args.embedding_mode == "ollama":
+            embedding_options["host"] = resolve_ollama_host(args.embedding_host)
+        elif args.embedding_mode == "openai":
+            embedding_options["base_url"] = resolve_openai_base_url(args.embedding_api_base)
+            resolved_embedding_key = resolve_openai_api_key(args.embedding_api_key)
+            if resolved_embedding_key:
+                embedding_options["api_key"] = resolved_embedding_key
+
         builder = LeannBuilder(
             backend_name=args.backend,
             embedding_model=args.embedding_model,
             embedding_mode=args.embedding_mode,
+            embedding_options=embedding_options or None,
             graph_degree=args.graph_degree,
             complexity=args.complexity,
             is_compact=args.compact,
@@ -1434,58 +1528,49 @@ Examples:
 
         llm_config = {"type": args.llm, "model": args.model}
         if args.llm == "ollama":
-            llm_config["host"] = args.host
+            llm_config["host"] = resolve_ollama_host(args.host)
+        elif args.llm == "openai":
+            llm_config["base_url"] = resolve_openai_base_url(args.api_base)
+            resolved_api_key = resolve_openai_api_key(args.api_key)
+            if resolved_api_key:
+                llm_config["api_key"] = resolved_api_key
 
         chat = LeannChat(index_path=index_path, llm_config=llm_config)
 
+        llm_kwargs: dict[str, Any] = {}
+        if args.thinking_budget:
+            llm_kwargs["thinking_budget"] = args.thinking_budget
+
+        def _ask_once(prompt: str) -> None:
+            response = chat.ask(
+                prompt,
+                top_k=args.top_k,
+                complexity=args.complexity,
+                beam_width=args.beam_width,
+                prune_ratio=args.prune_ratio,
+                recompute_embeddings=args.recompute_embeddings,
+                pruning_strategy=args.pruning_strategy,
+                llm_kwargs=llm_kwargs,
+            )
+            print(f"LEANN: {response}")
+
+        initial_query = (args.query or "").strip()
+
         if args.interactive:
-            print("LEANN Assistant ready! Type 'quit' to exit")
-            print("=" * 40)
+            # Create interactive session
+            session = create_cli_session(index_name)
 
-            while True:
-                user_input = input("\nYou: ").strip()
-                if user_input.lower() in ["quit", "exit", "q"]:
-                    print("Goodbye!")
-                    break
+            if initial_query:
+                _ask_once(initial_query)
 
-                if not user_input:
-                    continue
-
-                # Prepare LLM kwargs with thinking budget if specified
-                llm_kwargs = {}
-                if args.thinking_budget:
-                    llm_kwargs["thinking_budget"] = args.thinking_budget
-
-                response = chat.ask(
-                    user_input,
-                    top_k=args.top_k,
-                    complexity=args.complexity,
-                    beam_width=args.beam_width,
-                    prune_ratio=args.prune_ratio,
-                    recompute_embeddings=args.recompute_embeddings,
-                    pruning_strategy=args.pruning_strategy,
-                    llm_kwargs=llm_kwargs,
-                )
-                print(f"LEANN: {response}")
+            session.run_interactive_loop(_ask_once)
         else:
-            query = input("Enter your question: ").strip()
-            if query:
-                # Prepare LLM kwargs with thinking budget if specified
-                llm_kwargs = {}
-                if args.thinking_budget:
-                    llm_kwargs["thinking_budget"] = args.thinking_budget
+            query = initial_query or input("Enter your question: ").strip()
+            if not query:
+                print("No question provided. Exiting.")
+                return
 
-                response = chat.ask(
-                    query,
-                    top_k=args.top_k,
-                    complexity=args.complexity,
-                    beam_width=args.beam_width,
-                    prune_ratio=args.prune_ratio,
-                    recompute_embeddings=args.recompute_embeddings,
-                    pruning_strategy=args.pruning_strategy,
-                    llm_kwargs=llm_kwargs,
-                )
-                print(f"LEANN: {response}")
+            _ask_once(query)
 
     async def run(self, args=None):
         parser = self.create_parser()

packages/leann-core/src/leann/embedding_compute.py

@@ -7,11 +7,13 @@ Preserves all optimization parameters to ensure performance
 import logging
 import os
 import time
-from typing import Any
+from typing import Any, Optional
 
 import numpy as np
 import torch
 
+from .settings import resolve_ollama_host, resolve_openai_api_key, resolve_openai_base_url
+
 # Set up logger with proper level
 logger = logging.getLogger(__name__)
 LOG_LEVEL = os.getenv("LEANN_LOG_LEVEL", "WARNING").upper()
@@ -21,9 +23,6 @@ logger.setLevel(log_level)
 # Global model cache to avoid repeated loading
 _model_cache: dict[str, Any] = {}
 
-# Enable fast tokenizer multithreading by default
-os.environ.setdefault("TOKENIZERS_PARALLELISM", "true")
-
 
 def compute_embeddings(
     texts: list[str],
@@ -33,7 +32,8 @@ def compute_embeddings(
     batch_size: int = 32,
     adaptive_optimization: bool = True,
     manual_tokenize: bool = False,
-    max_length: int = 256,
+    max_length: int = 512,
+    provider_options: Optional[dict[str, Any]] = None,
 ) -> np.ndarray:
     """
     Unified embedding computation entry point
@@ -49,6 +49,8 @@ def compute_embeddings(
     Returns:
         Normalized embeddings array, shape: (len(texts), embedding_dim)
     """
+    provider_options = provider_options or {}
+
     if mode == "sentence-transformers":
         return compute_embeddings_sentence_transformers(
             texts,
@@ -60,11 +62,21 @@ def compute_embeddings(
             max_length=max_length,
         )
     elif mode == "openai":
-        return compute_embeddings_openai(texts, model_name)
+        return compute_embeddings_openai(
+            texts,
+            model_name,
+            base_url=provider_options.get("base_url"),
+            api_key=provider_options.get("api_key"),
+        )
     elif mode == "mlx":
         return compute_embeddings_mlx(texts, model_name)
     elif mode == "ollama":
-        return compute_embeddings_ollama(texts, model_name, is_build=is_build)
+        return compute_embeddings_ollama(
+            texts,
+            model_name,
+            is_build=is_build,
+            host=provider_options.get("host"),
+        )
     elif mode == "gemini":
         return compute_embeddings_gemini(texts, model_name, is_build=is_build)
     else:
@@ -80,11 +92,8 @@ def compute_embeddings_sentence_transformers(
     is_build: bool = False,
     adaptive_optimization: bool = True,
     manual_tokenize: bool = False,
-    max_length: int = 256,
+    max_length: int = 512,
 ) -> np.ndarray:
-    manual_tokenize = False
-    batch_size = 512
-    
     """
     Compute embeddings using SentenceTransformer with model caching and adaptive optimization
 
@@ -125,7 +134,7 @@ def compute_embeddings_sentence_transformers(
         # Keep original batch_size for CPU
 
     # Create cache key
-    cache_key = f"sentence_transformers_{model_name}_{device}_{use_fp16}_optimized_len{max_length}"
+    cache_key = f"sentence_transformers_{model_name}_{device}_{use_fp16}_optimized"
 
     # Check if model is already cached
     if cache_key in _model_cache:
@@ -164,18 +173,13 @@ def compute_embeddings_sentence_transformers(
             "torch_dtype": torch.float16 if use_fp16 else torch.float32,
             "low_cpu_mem_usage": True,
             "_fast_init": True,
+            "attn_implementation": "eager",  # Use eager attention for speed
         }
-        # Prefer SDPA on CUDA; fall back to eager elsewhere
-        if device == "cuda":
-            model_kwargs["attn_implementation"] = "sdpa"
-        else:
-            model_kwargs["attn_implementation"] = "eager"
 
         tokenizer_kwargs = {
             "use_fast": True,
-            "padding": "max_length",
+            "padding": True,
             "truncation": True,
-            "max_length": max_length,
         }
 
         try:
@@ -227,13 +231,6 @@ def compute_embeddings_sentence_transformers(
         for param in model.parameters():
             param.requires_grad_(False)
 
-        # Enforce max sequence length for encode path
-        try:
-            if hasattr(model, "max_seq_length"):
-                model.max_seq_length = max_length
-        except Exception:
-            pass
-
         # Cache the model
         _model_cache[cache_key] = model
         logger.info(f"Model cached: {cache_key}")
@@ -246,43 +243,22 @@ def compute_embeddings_sentence_transformers(
     start_time = time.time()
     if not manual_tokenize:
         # Use SentenceTransformer's optimized encode path (default)
-        # print text shapr
         with torch.inference_mode():
-            # print avg len of texts
-            avg_len = sum(len(text) for text in texts) / len(texts)
-            logger.info(f"Avg len of texts: {avg_len}")
-            # print the precision of the model
-            logger.info(f"Model precision: {model.dtype}")
-            time_start = time.time()
             embeddings = model.encode(
                 texts,
                 batch_size=batch_size,
                 show_progress_bar=is_build,  # Don't show progress bar in server environment
-                convert_to_tensor=True,
+                convert_to_numpy=True,
                 normalize_embeddings=False,
                 device=device,
-                max_length=max_length,
             )
-
         # Synchronize if CUDA to measure accurate wall time
         try:
-            # if torch.cuda.is_available():
-            #     torch.cuda.synchronize()
-            time_end = time.time()
-            embedding_time, embedding_tpt = (
-                time_end - time_start,
-                embeddings.shape[0] / (time_end - time_start),
-            )
-            logger.info(
-                f"Time taken in embedding {batch_size} texts in embedding model: {embedding_time} seconds, embedding tpt: {embedding_tpt} seqs/s"
-            )
+            if torch.cuda.is_available():
+                torch.cuda.synchronize()
         except Exception:
             pass
-        # Single CPU copy after timing (avoid per-batch D2H sync)
-        if isinstance(embeddings, torch.Tensor):
-            embeddings = embeddings.float().cpu().numpy()
     else:
-        time_start = time.time()
         # Manual tokenization + forward pass using HF AutoTokenizer/AutoModel
         try:
             from transformers import AutoModel, AutoTokenizer  # type: ignore
@@ -290,8 +266,8 @@ def compute_embeddings_sentence_transformers(
             raise ImportError(f"transformers is required for manual_tokenize=True: {e}")
 
         # Cache tokenizer and model
-        tok_cache_key = f"hf_tokenizer_{model_name}_len{max_length}_padmax"
-        mdl_cache_key = f"hf_model_{model_name}_{device}_{use_fp16}_len{max_length}"
+        tok_cache_key = f"hf_tokenizer_{model_name}"
+        mdl_cache_key = f"hf_model_{model_name}_{device}_{use_fp16}"
         if tok_cache_key in _model_cache and mdl_cache_key in _model_cache:
             hf_tokenizer = _model_cache[tok_cache_key]
             hf_model = _model_cache[mdl_cache_key]
@@ -312,10 +288,9 @@ def compute_embeddings_sentence_transformers(
             _model_cache[tok_cache_key] = hf_tokenizer
             _model_cache[mdl_cache_key] = hf_model
 
-        emb_list: list[torch.Tensor] = []
+        all_embeddings: list[np.ndarray] = []
         # Progress bar when building or for large inputs
         show_progress = is_build or len(texts) > 32
-        show_progress = False
         try:
             if show_progress:
                 from tqdm import tqdm  # type: ignore
@@ -338,36 +313,28 @@ def compute_embeddings_sentence_transformers(
                 tokenize_start_time = time.time()
                 inputs = hf_tokenizer(
                     batch_texts,
-                    padding="max_length",
+                    padding=True,
                     truncation=True,
                     max_length=max_length,
                     return_tensors="pt",
                 )
                 tokenize_end_time = time.time()
-                logger.debug(
+                logger.info(
                     f"Tokenize time taken: {tokenize_end_time - tokenize_start_time} seconds"
                 )
+                # Print shapes of all input tensors for debugging
+                for k, v in inputs.items():
+                    print(f"inputs[{k!r}] shape: {getattr(v, 'shape', type(v))}")
                 to_device_start_time = time.time()
-                # Pin CPU memory then transfer non-blocking to GPU when available
-                inputs = {
-                    k: (v.pin_memory() if (device == "cuda" and v.device.type == "cpu") else v)
-                    for k, v in inputs.items()
-                }
-                inputs = {
-                    k: v.to(device, non_blocking=(device == "cuda")) for k, v in inputs.items()
-                }
+                inputs = {k: v.to(device) for k, v in inputs.items()}
                 to_device_end_time = time.time()
-                logger.debug(
+                logger.info(
                     f"To device time taken: {to_device_end_time - to_device_start_time} seconds"
                 )
-                # if device == "cuda":
-                #     torch.cuda.synchronize()
                 forward_start_time = time.time()
                 outputs = hf_model(**inputs)
-                # if device == "cuda":
-                #     torch.cuda.synchronize()
                 forward_end_time = time.time()
-                logger.debug(f"Forward time taken: {forward_end_time - forward_start_time} seconds")
+                logger.info(f"Forward time taken: {forward_end_time - forward_start_time} seconds")
                 last_hidden_state = outputs.last_hidden_state  # (B, L, H)
                 attention_mask = inputs.get("attention_mask")
                 if attention_mask is None:
@@ -378,27 +345,18 @@ def compute_embeddings_sentence_transformers(
                     masked = last_hidden_state * mask
                     lengths = mask.sum(dim=1).clamp(min=1)
                     pooled = masked.sum(dim=1) / lengths
-                # Accumulate on-device; single D2H copy after loop
-                emb_list.append(pooled.detach())
+                # Move to CPU float32
+                batch_embeddings = pooled.detach().to("cpu").float().numpy()
+                all_embeddings.append(batch_embeddings)
 
-        # Concatenate and single-copy to CPU/NumPy
-        embeddings_tensor = torch.cat(emb_list, dim=0)
-        embeddings = embeddings_tensor.float().cpu().numpy()
-        # try:
-        #     if torch.cuda.is_available():
-        #         torch.cuda.synchronize()
-        # except Exception:
-        #     pass
+        embeddings = np.vstack(all_embeddings).astype(np.float32, copy=False)
+        try:
+            if torch.cuda.is_available():
+                torch.cuda.synchronize()
+        except Exception:
+            pass
         end_time = time.time()
         logger.info(f"Manual tokenize time taken: {end_time - start_time_manual} seconds")
-        time_end = time.time()
-        tokenize_time, tokenize_tpt = (
-            time_end - time_start,
-            embeddings.shape[0] / (time_end - time_start),
-        )
-        logger.info(
-            f"Tokenize time taken: {tokenize_time} seconds, tokenize tpt: {tokenize_tpt} seqs/s"
-        )
     end_time = time.time()
     logger.info(f"Generated {len(embeddings)} embeddings, dimension: {embeddings.shape[1]}")
     logger.info(f"Time taken: {end_time - start_time} seconds")
@@ -410,12 +368,15 @@ def compute_embeddings_sentence_transformers(
     return embeddings
 
 
-def compute_embeddings_openai(texts: list[str], model_name: str) -> np.ndarray:
+def compute_embeddings_openai(
+    texts: list[str],
+    model_name: str,
+    base_url: Optional[str] = None,
+    api_key: Optional[str] = None,
+) -> np.ndarray:
     # TODO: @yichuan-w add progress bar only in build mode
     """Compute embeddings using OpenAI API"""
     try:
-        import os
-
         import openai
     except ImportError as e:
         raise ImportError(f"OpenAI package not installed: {e}")
@@ -430,16 +391,18 @@ def compute_embeddings_openai(texts: list[str], model_name: str) -> np.ndarray:
             f"Found {invalid_count} empty/invalid text(s) in input. Upstream should filter before calling OpenAI."
         )
 
-    api_key = os.getenv("OPENAI_API_KEY")
-    if not api_key:
+    resolved_base_url = resolve_openai_base_url(base_url)
+    resolved_api_key = resolve_openai_api_key(api_key)
+
+    if not resolved_api_key:
         raise RuntimeError("OPENAI_API_KEY environment variable not set")
 
     # Cache OpenAI client
-    cache_key = "openai_client"
+    cache_key = f"openai_client::{resolved_base_url}"
     if cache_key in _model_cache:
         client = _model_cache[cache_key]
     else:
-        client = openai.OpenAI(api_key=api_key)
+        client = openai.OpenAI(api_key=resolved_api_key, base_url=resolved_base_url)
         _model_cache[cache_key] = client
         logger.info("OpenAI client cached")
 
@@ -564,7 +527,10 @@ def compute_embeddings_mlx(chunks: list[str], model_name: str, batch_size: int =
 
 
 def compute_embeddings_ollama(
-    texts: list[str], model_name: str, is_build: bool = False, host: str = "http://localhost:11434"
+    texts: list[str],
+    model_name: str,
+    is_build: bool = False,
+    host: Optional[str] = None,
 ) -> np.ndarray:
     """
     Compute embeddings using Ollama API with simplified batch processing.
@@ -575,7 +541,7 @@ def compute_embeddings_ollama(
         texts: List of texts to compute embeddings for
         model_name: Ollama model name (e.g., "nomic-embed-text", "mxbai-embed-large")
         is_build: Whether this is a build operation (shows progress bar)
-        host: Ollama host URL (default: http://localhost:11434)
+        host: Ollama host URL (defaults to environment or http://localhost:11434)
 
     Returns:
         Normalized embeddings array, shape: (len(texts), embedding_dim)
@@ -590,17 +556,19 @@ def compute_embeddings_ollama(
     if not texts:
         raise ValueError("Cannot compute embeddings for empty text list")
 
+    resolved_host = resolve_ollama_host(host)
+
     logger.info(
-        f"Computing embeddings for {len(texts)} texts using Ollama API, model: '{model_name}'"
+        f"Computing embeddings for {len(texts)} texts using Ollama API, model: '{model_name}', host: '{resolved_host}'"
     )
 
     # Check if Ollama is running
     try:
-        response = requests.get(f"{host}/api/version", timeout=5)
+        response = requests.get(f"{resolved_host}/api/version", timeout=5)
         response.raise_for_status()
     except requests.exceptions.ConnectionError:
         error_msg = (
-            f"❌ Could not connect to Ollama at {host}.\n\n"
+            f"❌ Could not connect to Ollama at {resolved_host}.\n\n"
             "Please ensure Ollama is running:\n"
             "  • macOS/Linux: ollama serve\n"
             "  • Windows: Make sure Ollama is running in the system tray\n\n"
@@ -612,7 +580,7 @@ def compute_embeddings_ollama(
 
     # Check if model exists and provide helpful suggestions
     try:
-        response = requests.get(f"{host}/api/tags", timeout=5)
+        response = requests.get(f"{resolved_host}/api/tags", timeout=5)
         response.raise_for_status()
         models = response.json()
         model_names = [model["name"] for model in models.get("models", [])]
@@ -675,7 +643,9 @@ def compute_embeddings_ollama(
         # Verify the model supports embeddings by testing it
         try:
             test_response = requests.post(
-                f"{host}/api/embeddings", json={"model": model_name, "prompt": "test"}, timeout=10
+                f"{resolved_host}/api/embeddings",
+                json={"model": model_name, "prompt": "test"},
+                timeout=10,
             )
             if test_response.status_code != 200:
                 error_msg = (
@@ -722,7 +692,7 @@ def compute_embeddings_ollama(
             while retry_count < max_retries:
                 try:
                     response = requests.post(
-                        f"{host}/api/embeddings",
+                        f"{resolved_host}/api/embeddings",
                         json={"model": model_name, "prompt": truncated_text},
                         timeout=30,
                     )

packages/leann-core/src/leann/embedding_server_manager.py

@@ -1,4 +1,5 @@
 import atexit
+import json
 import logging
 import os
 import socket
@@ -8,6 +9,8 @@ import time
 from pathlib import Path
 from typing import Optional
 
+from .settings import encode_provider_options
+
 # Lightweight, self-contained server manager with no cross-process inspection
 
 # Set up logging based on environment variable
@@ -46,6 +49,85 @@ def _check_port(port: int) -> bool:
 # Note: All cross-process scanning helpers removed for simplicity
 
 
+def _safe_resolve(path: Path) -> str:
+    """Resolve paths safely even if the target does not yet exist."""
+    try:
+        return str(path.resolve(strict=False))
+    except Exception:
+        return str(path)
+
+
+def _safe_stat_signature(path: Path) -> dict:
+    """Return a lightweight signature describing the current state of a path."""
+    signature: dict[str, object] = {"path": _safe_resolve(path)}
+    try:
+        stat = path.stat()
+    except FileNotFoundError:
+        signature["missing"] = True
+    except Exception as exc:  # pragma: no cover - unexpected filesystem errors
+        signature["error"] = str(exc)
+    else:
+        signature["mtime_ns"] = stat.st_mtime_ns
+        signature["size"] = stat.st_size
+    return signature
+
+
+def _build_passages_signature(passages_file: Optional[str]) -> Optional[dict]:
+    """Collect modification signatures for metadata and referenced passage files."""
+    if not passages_file:
+        return None
+
+    meta_path = Path(passages_file)
+    signature: dict[str, object] = {"meta": _safe_stat_signature(meta_path)}
+
+    try:
+        with meta_path.open(encoding="utf-8") as fh:
+            meta = json.load(fh)
+    except FileNotFoundError:
+        signature["meta_missing"] = True
+        signature["sources"] = []
+        return signature
+    except json.JSONDecodeError as exc:
+        signature["meta_error"] = f"json_error:{exc}"
+        signature["sources"] = []
+        return signature
+    except Exception as exc:  # pragma: no cover - unexpected errors
+        signature["meta_error"] = str(exc)
+        signature["sources"] = []
+        return signature
+
+    base_dir = meta_path.parent
+    seen_paths: set[str] = set()
+    source_signatures: list[dict[str, object]] = []
+
+    for source in meta.get("passage_sources", []):
+        for key, kind in (
+            ("path", "passages"),
+            ("path_relative", "passages"),
+            ("index_path", "index"),
+            ("index_path_relative", "index"),
+        ):
+            raw_path = source.get(key)
+            if not raw_path:
+                continue
+            candidate = Path(raw_path)
+            if not candidate.is_absolute():
+                candidate = base_dir / candidate
+            resolved = _safe_resolve(candidate)
+            if resolved in seen_paths:
+                continue
+            seen_paths.add(resolved)
+            sig = _safe_stat_signature(candidate)
+            sig["kind"] = kind
+            source_signatures.append(sig)
+
+    signature["sources"] = source_signatures
+    return signature
+
+
+# Note: All cross-process scanning helpers removed for simplicity
+
+
 class EmbeddingServerManager:
     """
     A simplified manager for embedding server processes that avoids complex update mechanisms.
@@ -82,16 +164,42 @@ class EmbeddingServerManager:
     ) -> tuple[bool, int]:
         """Start the embedding server."""
         # passages_file may be present in kwargs for server CLI, but we don't need it here
+        provider_options = kwargs.pop("provider_options", None)
+        passages_file = kwargs.get("passages_file", "")
+
+        config_signature = self._build_config_signature(
+            model_name=model_name,
+            embedding_mode=embedding_mode,
+            provider_options=provider_options,
+            passages_file=passages_file,
+        )
 
         # If this manager already has a live server, just reuse it
-        if self.server_process and self.server_process.poll() is None and self.server_port:
+        if (
+            self.server_process
+            and self.server_process.poll() is None
+            and self.server_port
+            and self._server_config == config_signature
+        ):
             logger.info("Reusing in-process server")
             return True, self.server_port
 
+        # Configuration changed, stop existing server before starting a new one
+        if self.server_process and self.server_process.poll() is None:
+            logger.info("Existing server configuration differs; restarting embedding server")
+            self.stop_server()
+
         # For Colab environment, use a different strategy
         if _is_colab_environment():
             logger.info("Detected Colab environment, using alternative startup strategy")
-            return self._start_server_colab(port, model_name, embedding_mode, **kwargs)
+            return self._start_server_colab(
+                port,
+                model_name,
+                embedding_mode,
+                config_signature=config_signature,
+                provider_options=provider_options,
+                **kwargs,
+            )
 
         # Always pick a fresh available port
         try:
@@ -101,13 +209,40 @@ class EmbeddingServerManager:
             return False, port
 
         # Start a new server
-        return self._start_new_server(actual_port, model_name, embedding_mode, **kwargs)
+        return self._start_new_server(
+            actual_port,
+            model_name,
+            embedding_mode,
+            provider_options=provider_options,
+            config_signature=config_signature,
+            **kwargs,
+        )
+
+    def _build_config_signature(
+        self,
+        *,
+        model_name: str,
+        embedding_mode: str,
+        provider_options: Optional[dict],
+        passages_file: Optional[str],
+    ) -> dict:
+        """Create a signature describing the current server configuration."""
+        return {
+            "model_name": model_name,
+            "passages_file": passages_file or "",
+            "embedding_mode": embedding_mode,
+            "provider_options": provider_options or {},
+            "passages_signature": _build_passages_signature(passages_file),
+        }
 
     def _start_server_colab(
         self,
         port: int,
         model_name: str,
         embedding_mode: str = "sentence-transformers",
+        *,
+        config_signature: Optional[dict] = None,
+        provider_options: Optional[dict] = None,
         **kwargs,
     ) -> tuple[bool, int]:
         """Start server with Colab-specific configuration."""
@@ -125,8 +260,21 @@ class EmbeddingServerManager:
 
         try:
             # In Colab, we'll use a more direct approach
-            self._launch_server_process_colab(command, actual_port)
-            return self._wait_for_server_ready_colab(actual_port)
+            self._launch_server_process_colab(
+                command,
+                actual_port,
+                provider_options=provider_options,
+                config_signature=config_signature,
+            )
+            started, ready_port = self._wait_for_server_ready_colab(actual_port)
+            if started:
+                self._server_config = config_signature or {
+                    "model_name": model_name,
+                    "passages_file": kwargs.get("passages_file", ""),
+                    "embedding_mode": embedding_mode,
+                    "provider_options": provider_options or {},
+                }
+            return started, ready_port
         except Exception as e:
             logger.error(f"Failed to start embedding server in Colab: {e}")
             return False, actual_port
@@ -134,7 +282,13 @@ class EmbeddingServerManager:
     # Note: No compatibility check needed; manager is per-searcher and configs are stable per instance
 
     def _start_new_server(
-        self, port: int, model_name: str, embedding_mode: str, **kwargs
+        self,
+        port: int,
+        model_name: str,
+        embedding_mode: str,
+        provider_options: Optional[dict] = None,
+        config_signature: Optional[dict] = None,
+        **kwargs,
     ) -> tuple[bool, int]:
         """Start a new embedding server on the given port."""
         logger.info(f"Starting embedding server on port {port}...")
@@ -142,8 +296,21 @@ class EmbeddingServerManager:
         command = self._build_server_command(port, model_name, embedding_mode, **kwargs)
 
         try:
-            self._launch_server_process(command, port)
-            return self._wait_for_server_ready(port)
+            self._launch_server_process(
+                command,
+                port,
+                provider_options=provider_options,
+                config_signature=config_signature,
+            )
+            started, ready_port = self._wait_for_server_ready(port)
+            if started:
+                self._server_config = config_signature or {
+                    "model_name": model_name,
+                    "passages_file": kwargs.get("passages_file", ""),
+                    "embedding_mode": embedding_mode,
+                    "provider_options": provider_options or {},
+                }
+            return started, ready_port
         except Exception as e:
             logger.error(f"Failed to start embedding server: {e}")
             return False, port
@@ -173,7 +340,14 @@ class EmbeddingServerManager:
 
         return command
 
-    def _launch_server_process(self, command: list, port: int) -> None:
+    def _launch_server_process(
+        self,
+        command: list,
+        port: int,
+        *,
+        provider_options: Optional[dict] = None,
+        config_signature: Optional[dict] = None,
+    ) -> None:
         """Launch the server process."""
         project_root = Path(__file__).parent.parent.parent.parent.parent
         logger.info(f"Command: {' '.join(command)}")
@@ -193,14 +367,23 @@ class EmbeddingServerManager:
 
         # Start embedding server subprocess
         logger.info(f"Starting server process with command: {' '.join(command)}")
+        env = os.environ.copy()
+        encoded_options = encode_provider_options(provider_options)
+        if encoded_options:
+            env["LEANN_EMBEDDING_OPTIONS"] = encoded_options
+
         self.server_process = subprocess.Popen(
             command,
             cwd=project_root,
             stdout=stdout_target,
             stderr=stderr_target,
+            env=env,
         )
         self.server_port = port
-        # Record config for in-process reuse
+        # Record config for in-process reuse (best effort; refined later when ready)
+        if config_signature is not None:
+            self._server_config = config_signature
+        else:  # Fallback for unexpected code paths
             try:
                 self._server_config = {
                     "model_name": command[command.index("--model-name") + 1]
@@ -212,12 +395,14 @@ class EmbeddingServerManager:
                     "embedding_mode": command[command.index("--embedding-mode") + 1]
                     if "--embedding-mode" in command
                     else "sentence-transformers",
+                    "provider_options": provider_options or {},
                 }
             except Exception:
                 self._server_config = {
                     "model_name": "",
                     "passages_file": "",
                     "embedding_mode": "sentence-transformers",
+                    "provider_options": provider_options or {},
                 }
         logger.info(f"Server process started with PID: {self.server_process.pid}")
 
@@ -322,16 +507,29 @@ class EmbeddingServerManager:
         # Removed: cross-process adoption no longer supported
         return
 
-    def _launch_server_process_colab(self, command: list, port: int) -> None:
+    def _launch_server_process_colab(
+        self,
+        command: list,
+        port: int,
+        *,
+        provider_options: Optional[dict] = None,
+        config_signature: Optional[dict] = None,
+    ) -> None:
         """Launch the server process with Colab-specific settings."""
         logger.info(f"Colab Command: {' '.join(command)}")
 
         # In Colab, we need to be more careful about process management
+        env = os.environ.copy()
+        encoded_options = encode_provider_options(provider_options)
+        if encoded_options:
+            env["LEANN_EMBEDDING_OPTIONS"] = encoded_options
+
         self.server_process = subprocess.Popen(
             command,
             stdout=subprocess.PIPE,
             stderr=subprocess.PIPE,
             text=True,
+            env=env,
         )
         self.server_port = port
         logger.info(f"Colab server process started with PID: {self.server_process.pid}")
@@ -341,10 +539,14 @@ class EmbeddingServerManager:
             atexit.register(self._finalize_process)
             self._atexit_registered = True
         # Record config for in-process reuse is best-effort in Colab mode
+        if config_signature is not None:
+            self._server_config = config_signature
+        else:
             self._server_config = {
                 "model_name": "",
                 "passages_file": "",
                 "embedding_mode": "sentence-transformers",
+                "provider_options": provider_options or {},
             }
 
     def _wait_for_server_ready_colab(self, port: int) -> tuple[bool, int]:

packages/leann-core/src/leann/interactive_utils.py

@@ -0,0 +1,189 @@
+"""
+Interactive session utilities for LEANN applications.
+
+Provides shared readline functionality and command handling across
+CLI, API, and RAG example interactive modes.
+"""
+
+import atexit
+import os
+from pathlib import Path
+from typing import Callable, Optional
+
+# Try to import readline with fallback for Windows
+try:
+    import readline
+
+    HAS_READLINE = True
+except ImportError:
+    # Windows doesn't have readline by default
+    HAS_READLINE = False
+    readline = None
+
+
+class InteractiveSession:
+    """Manages interactive session with optional readline support and common commands."""
+
+    def __init__(
+        self,
+        history_name: str,
+        prompt: str = "You: ",
+        welcome_message: str = "",
+    ):
+        """
+        Initialize interactive session with optional readline support.
+
+        Args:
+            history_name: Name for history file (e.g., "cli", "api_chat")
+                         (ignored if readline not available)
+            prompt: Input prompt to display
+            welcome_message: Message to show when starting session
+
+        Note:
+            On systems without readline (e.g., Windows), falls back to basic input()
+            with limited functionality (no history, no line editing).
+        """
+        self.history_name = history_name
+        self.prompt = prompt
+        self.welcome_message = welcome_message
+        self._setup_complete = False
+
+    def setup_readline(self):
+        """Setup readline with history support (if available)."""
+        if self._setup_complete:
+            return
+
+        if not HAS_READLINE:
+            # Readline not available (likely Windows), skip setup
+            self._setup_complete = True
+            return
+
+        # History file setup
+        history_dir = Path.home() / ".leann" / "history"
+        history_dir.mkdir(parents=True, exist_ok=True)
+        history_file = history_dir / f"{self.history_name}.history"
+
+        # Load history if exists
+        try:
+            readline.read_history_file(str(history_file))
+            readline.set_history_length(1000)
+        except FileNotFoundError:
+            pass
+
+        # Save history on exit
+        atexit.register(readline.write_history_file, str(history_file))
+
+        # Optional: Enable vi editing mode (commented out by default)
+        # readline.parse_and_bind("set editing-mode vi")
+
+        self._setup_complete = True
+
+    def _show_help(self):
+        """Show available commands."""
+        print("Commands:")
+        print("  quit/exit/q - Exit the chat")
+        print("  help - Show this help message")
+        print("  clear - Clear screen")
+        print("  history - Show command history")
+
+    def _show_history(self):
+        """Show command history."""
+        if not HAS_READLINE:
+            print("  History not available (readline not supported on this system)")
+            return
+
+        history_length = readline.get_current_history_length()
+        if history_length == 0:
+            print("  No history available")
+            return
+
+        for i in range(history_length):
+            item = readline.get_history_item(i + 1)
+            if item:
+                print(f"  {i + 1}: {item}")
+
+    def get_user_input(self) -> Optional[str]:
+        """
+        Get user input with readline support.
+
+        Returns:
+            User input string, or None if EOF (Ctrl+D)
+        """
+        try:
+            return input(self.prompt).strip()
+        except KeyboardInterrupt:
+            print("\n(Use 'quit' to exit)")
+            return ""  # Return empty string to continue
+        except EOFError:
+            print("\nGoodbye!")
+            return None
+
+    def run_interactive_loop(self, handler_func: Callable[[str], None]):
+        """
+        Run the interactive loop with a custom handler function.
+
+        Args:
+            handler_func: Function to handle user input that's not a built-in command
+                         Should accept a string and handle the user's query
+        """
+        self.setup_readline()
+
+        if self.welcome_message:
+            print(self.welcome_message)
+
+        while True:
+            user_input = self.get_user_input()
+
+            if user_input is None:  # EOF (Ctrl+D)
+                break
+
+            if not user_input:  # Empty input or KeyboardInterrupt
+                continue
+
+            # Handle built-in commands
+            command = user_input.lower()
+            if command in ["quit", "exit", "q"]:
+                print("Goodbye!")
+                break
+            elif command == "help":
+                self._show_help()
+            elif command == "clear":
+                os.system("clear" if os.name != "nt" else "cls")
+            elif command == "history":
+                self._show_history()
+            else:
+                # Regular user input - pass to handler
+                try:
+                    handler_func(user_input)
+                except Exception as e:
+                    print(f"Error: {e}")
+
+
+def create_cli_session(index_name: str) -> InteractiveSession:
+    """Create an interactive session for CLI usage."""
+    return InteractiveSession(
+        history_name=index_name,
+        prompt="\nYou: ",
+        welcome_message="LEANN Assistant ready! Type 'quit' to exit, 'help' for commands\n"
+        + "=" * 40,
+    )
+
+
+def create_api_session() -> InteractiveSession:
+    """Create an interactive session for API chat."""
+    return InteractiveSession(
+        history_name="api_chat",
+        prompt="You: ",
+        welcome_message="Leann Chat started (type 'quit' to exit, 'help' for commands)\n"
+        + "=" * 40,
+    )
+
+
+def create_rag_session(app_name: str, data_description: str) -> InteractiveSession:
+    """Create an interactive session for RAG examples."""
+    return InteractiveSession(
+        history_name=f"{app_name}_rag",
+        prompt="You: ",
+        welcome_message=f"[Interactive Mode] Chat with your {data_description} data!\nType 'quit' or 'exit' to stop, 'help' for commands.\n"
+        + "=" * 40,
+    )

packages/leann-core/src/leann/searcher_base.py

@@ -41,6 +41,7 @@ class BaseSearcher(LeannBackendSearcherInterface, ABC):
             print("WARNING: embedding_model not found in meta.json. Recompute will fail.")
 
         self.embedding_mode = self.meta.get("embedding_mode", "sentence-transformers")
+        self.embedding_options = self.meta.get("embedding_options", {})
 
         self.embedding_server_manager = EmbeddingServerManager(
             backend_module_name=backend_module_name,
@@ -77,6 +78,7 @@ class BaseSearcher(LeannBackendSearcherInterface, ABC):
             passages_file=passages_source_file,
             distance_metric=distance_metric,
             enable_warmup=kwargs.get("enable_warmup", False),
+            provider_options=self.embedding_options,
         )
         if not server_started:
             raise RuntimeError(f"Failed to start embedding server on port {actual_port}")
@@ -125,7 +127,12 @@ class BaseSearcher(LeannBackendSearcherInterface, ABC):
         from .embedding_compute import compute_embeddings
 
         embedding_mode = self.meta.get("embedding_mode", "sentence-transformers")
-        return compute_embeddings([query], self.embedding_model, embedding_mode)
+        return compute_embeddings(
+            [query],
+            self.embedding_model,
+            embedding_mode,
+            provider_options=self.embedding_options,
+        )
 
     def _compute_embedding_via_server(self, chunks: list, zmq_port: int) -> np.ndarray:
         """Compute embeddings using the ZMQ embedding server."""

packages/leann-core/src/leann/settings.py

@@ -0,0 +1,74 @@
+"""Runtime configuration helpers for LEANN."""
+
+from __future__ import annotations
+
+import json
+import os
+from typing import Any
+
+# Default fallbacks to preserve current behaviour while keeping them in one place.
+_DEFAULT_OLLAMA_HOST = "http://localhost:11434"
+_DEFAULT_OPENAI_BASE_URL = "https://api.openai.com/v1"
+
+
+def _clean_url(value: str) -> str:
+    """Normalize URL strings by stripping trailing slashes."""
+
+    return value.rstrip("/") if value else value
+
+
+def resolve_ollama_host(explicit: str | None = None) -> str:
+    """Resolve the Ollama-compatible endpoint to use."""
+
+    candidates = (
+        explicit,
+        os.getenv("LEANN_LOCAL_LLM_HOST"),
+        os.getenv("LEANN_OLLAMA_HOST"),
+        os.getenv("OLLAMA_HOST"),
+        os.getenv("LOCAL_LLM_ENDPOINT"),
+    )
+
+    for candidate in candidates:
+        if candidate:
+            return _clean_url(candidate)
+
+    return _clean_url(_DEFAULT_OLLAMA_HOST)
+
+
+def resolve_openai_base_url(explicit: str | None = None) -> str:
+    """Resolve the base URL for OpenAI-compatible services."""
+
+    candidates = (
+        explicit,
+        os.getenv("LEANN_OPENAI_BASE_URL"),
+        os.getenv("OPENAI_BASE_URL"),
+        os.getenv("LOCAL_OPENAI_BASE_URL"),
+    )
+
+    for candidate in candidates:
+        if candidate:
+            return _clean_url(candidate)
+
+    return _clean_url(_DEFAULT_OPENAI_BASE_URL)
+
+
+def resolve_openai_api_key(explicit: str | None = None) -> str | None:
+    """Resolve the API key for OpenAI-compatible services."""
+
+    if explicit:
+        return explicit
+
+    return os.getenv("OPENAI_API_KEY")
+
+
+def encode_provider_options(options: dict[str, Any] | None) -> str | None:
+    """Serialize provider options for child processes."""
+
+    if not options:
+        return None
+
+    try:
+        return json.dumps(options)
+    except (TypeError, ValueError):
+        # Fall back to empty payload if serialization fails
+        return None

packages/leann-mcp/README.md

@@ -2,6 +2,8 @@
 
 Transform your development workflow with intelligent code assistance using LEANN's semantic search directly in Claude Code.
 
+For agent-facing discovery details, see `llms.txt` in the repository root.
+
 ## Prerequisites
 
 Install LEANN globally for MCP integration (with default backend):

packages/leann/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "leann"
-version = "0.3.2"
+version = "0.3.4"
 description = "LEANN - The smallest vector index in the world. RAG Everything with LEANN!"
 readme = "README.md"
 requires-python = ">=3.9"

pyproject.toml

@@ -22,7 +22,10 @@ dependencies = [
     "sglang",
     "ollama",
     "requests>=2.25.0",
-    "sentence-transformers>=2.2.0",
+    "sentence-transformers>=3.0.0",
+    # Pin transformers below 4.46: 4.46.0 introduced Python 3.10-only typing (PEP 604) and
+    # breaks our Python 3.9 test matrix when pulled in by sentence-transformers.
+    "transformers<4.46",
     "openai>=1.0.0",
     # PDF parsing dependencies - essential for document processing
     "PyPDF2>=3.0.0",
@@ -53,27 +56,10 @@ dependencies = [
     "tree-sitter-java>=0.20.0",
     "tree-sitter-c-sharp>=0.20.0",
     "tree-sitter-typescript>=0.20.0",
+    "torchvision>=0.23.0",
 ]
 
 [project.optional-dependencies]
-dev = [
-    "pytest>=7.0",
-    "pytest-cov>=4.0",
-    "pytest-xdist>=3.0",  # For parallel test execution
-    "black>=23.0",
-    "ruff==0.12.7",  # Fixed version to ensure consistent formatting across all environments
-    "matplotlib",
-    "huggingface-hub>=0.20.0",
-    "pre-commit>=3.5.0",
-]
-
-test = [
-    "pytest>=7.0",
-    "pytest-timeout>=2.0",
-    "llama-index-core>=0.12.0",
-    "python-dotenv>=1.0.0",
-]
-
 diskann = [
     "leann-backend-diskann",
 ]
@@ -99,11 +85,38 @@ wechat-exporter = "wechat_exporter.main:main"
 leann-core = { path = "packages/leann-core", editable = true }
 leann-backend-diskann = { path = "packages/leann-backend-diskann", editable = true }
 leann-backend-hnsw = { path = "packages/leann-backend-hnsw", editable = true }
+astchunk = { path = "packages/astchunk-leann", editable = true }
+
+[dependency-groups]
+# Minimal lint toolchain for CI and local hooks
+lint = [
+    "pre-commit>=3.5.0",
+    "ruff==0.12.7",  # Fixed version to ensure consistent formatting across all environments
+]
+
+# Test toolchain (no heavy project runtime deps)
+test = [
+    "pytest>=7.0",
+    "pytest-cov>=4.0",
+    "pytest-xdist>=3.0",
+    "pytest-timeout>=2.0",
+    "python-dotenv>=1.0.0",
+]
+
+# dependencies by apps/ should list here
+dev = [
+    "matplotlib",
+    "huggingface-hub>=0.20.0",
+]
 
 [tool.ruff]
 target-version = "py39"
 line-length = 100
-extend-exclude = ["third_party"]
+extend-exclude = [
+    "third_party",
+    "apps/multimodal/vision-based-pdf-multi-vector/multi-vector-leann-paper-example.py",
+    "apps/multimodal/vision-based-pdf-multi-vector/multi-vector-leann-similarity-map.py"
+]
 
 
 [tool.ruff.lint]

scripts/hf_upload.py

@@ -0,0 +1,121 @@
+#!/usr/bin/env python3
+"""
+Upload local evaluation data to Hugging Face Hub, excluding diskann_rpj_wiki.
+
+Defaults:
+- repo_id: LEANN-RAG/leann-rag-evaluation-data (dataset)
+- folder_path: benchmarks/data
+- ignore_patterns: diskann_rpj_wiki/** and .cache/**
+
+Requires authentication via `huggingface-cli login` or HF_TOKEN env var.
+"""
+
+from __future__ import annotations
+
+import argparse
+import os
+
+try:
+    from huggingface_hub import HfApi
+except Exception as e:
+    raise SystemExit(
+        "huggingface_hub is required. Install with: pip install huggingface_hub hf_transfer"
+    ) from e
+
+
+def _enable_transfer_accel_if_available() -> None:
+    """Best-effort enabling of accelerated transfers across hub versions.
+
+    Tries the public util if present; otherwise, falls back to env flag when
+    hf_transfer is installed. Silently no-ops if unavailable.
+    """
+    try:
+        # Newer huggingface_hub exposes this under utils
+        from huggingface_hub.utils import hf_hub_enable_hf_transfer  # type: ignore
+
+        hf_hub_enable_hf_transfer()
+        return
+    except Exception:
+        pass
+
+    try:
+        # If hf_transfer is installed, set env flag recognized by the hub
+        import hf_transfer  # noqa: F401
+
+        os.environ.setdefault("HF_HUB_ENABLE_HF_TRANSFER", "1")
+    except Exception:
+        # Acceleration not available; proceed without it
+        pass
+
+
+def parse_args() -> argparse.Namespace:
+    p = argparse.ArgumentParser(description="Upload local data to HF, excluding diskann_rpj_wiki")
+    p.add_argument(
+        "--repo-id",
+        default="LEANN-RAG/leann-rag-evaluation-data",
+        help="Target dataset repo id (namespace/name)",
+    )
+    p.add_argument(
+        "--folder-path",
+        default="benchmarks/data",
+        help="Local folder to upload (default: benchmarks/data)",
+    )
+    p.add_argument(
+        "--ignore",
+        default=["diskann_rpj_wiki/**", ".cache/**"],
+        nargs="+",
+        help="Glob patterns to ignore (space-separated)",
+    )
+    p.add_argument(
+        "--allow",
+        default=["**"],
+        nargs="+",
+        help="Glob patterns to allow (space-separated). Defaults to everything.",
+    )
+    p.add_argument(
+        "--message",
+        default="sync local data (exclude diskann_rpj_wiki)",
+        help="Commit message",
+    )
+    p.add_argument(
+        "--no-transfer-accel",
+        action="store_true",
+        help="Disable hf_transfer accelerated uploads",
+    )
+    return p.parse_args()
+
+
+def main() -> None:
+    args = parse_args()
+
+    if not args.no_transfer_accel:
+        _enable_transfer_accel_if_available()
+
+    if not os.path.isdir(args.folder_path):
+        raise SystemExit(f"Folder not found: {args.folder_path}")
+
+    print("Uploading to Hugging Face Hub:")
+    print(f"  repo_id:        {args.repo_id}")
+    print("  repo_type:      dataset")
+    print(f"  folder_path:    {args.folder_path}")
+    print(f"  allow_patterns: {args.allow}")
+    print(f"  ignore_patterns:{args.ignore}")
+
+    api = HfApi()
+
+    # Perform upload. This skips unchanged files by content hash.
+    api.upload_folder(
+        repo_id=args.repo_id,
+        repo_type="dataset",
+        folder_path=args.folder_path,
+        path_in_repo=".",
+        allow_patterns=args.allow,
+        ignore_patterns=args.ignore,
+        commit_message=args.message,
+    )
+
+    print("Upload completed (unchanged files were skipped by the Hub).")
+
+
+if __name__ == "__main__":
+    main()

tests/README.md

@@ -40,8 +40,8 @@ Tests DiskANN graph partitioning functionality:
 
 ### Install test dependencies:
 ```bash
-# Using extras
-uv pip install -e ".[test]"
+# Using uv dependency groups (tools only)
+uv sync --only-group test
 ```
 
 ### Run all tests:

tests/test_cli_ask.py

@@ -0,0 +1,14 @@
+from leann.cli import LeannCLI
+
+
+def test_cli_ask_accepts_positional_query(tmp_path, monkeypatch):
+    monkeypatch.chdir(tmp_path)
+
+    cli = LeannCLI()
+    parser = cli.create_parser()
+
+    args = parser.parse_args(["ask", "my-docs", "Where are prompts configured?"])
+
+    assert args.command == "ask"
+    assert args.index_name == "my-docs"
+    assert args.query == "Where are prompts configured?"

tests/test_embedding_server_manager.py

@@ -0,0 +1,137 @@
+import json
+import time
+
+import pytest
+from leann.embedding_server_manager import EmbeddingServerManager
+
+
+class DummyProcess:
+    def __init__(self):
+        self.pid = 12345
+        self._terminated = False
+
+    def poll(self):
+        return 0 if self._terminated else None
+
+    def terminate(self):
+        self._terminated = True
+
+    def kill(self):
+        self._terminated = True
+
+    def wait(self, timeout=None):
+        self._terminated = True
+        return 0
+
+
+@pytest.fixture
+def embedding_manager(monkeypatch):
+    manager = EmbeddingServerManager("leann_backend_hnsw.hnsw_embedding_server")
+
+    def fake_get_available_port(start_port):
+        return start_port
+
+    monkeypatch.setattr(
+        "leann.embedding_server_manager._get_available_port",
+        fake_get_available_port,
+    )
+
+    start_calls = []
+
+    def fake_start_new_server(self, port, model_name, embedding_mode, **kwargs):
+        config_signature = kwargs.get("config_signature")
+        start_calls.append(config_signature)
+        self.server_process = DummyProcess()
+        self.server_port = port
+        self._server_config = config_signature
+        return True, port
+
+    monkeypatch.setattr(
+        EmbeddingServerManager,
+        "_start_new_server",
+        fake_start_new_server,
+    )
+
+    # Ensure stop_server doesn't try to operate on real subprocesses
+    def fake_stop_server(self):
+        self.server_process = None
+        self.server_port = None
+        self._server_config = None
+
+    monkeypatch.setattr(EmbeddingServerManager, "stop_server", fake_stop_server)
+
+    return manager, start_calls
+
+
+def _write_meta(meta_path, passages_name, index_name, total):
+    meta_path.write_text(
+        json.dumps(
+            {
+                "backend_name": "hnsw",
+                "embedding_model": "test-model",
+                "embedding_mode": "sentence-transformers",
+                "dimensions": 3,
+                "backend_kwargs": {},
+                "passage_sources": [
+                    {
+                        "type": "jsonl",
+                        "path": passages_name,
+                        "index_path": index_name,
+                    }
+                ],
+                "total_passages": total,
+            }
+        ),
+        encoding="utf-8",
+    )
+
+
+def test_server_restarts_when_metadata_changes(tmp_path, embedding_manager):
+    manager, start_calls = embedding_manager
+
+    meta_path = tmp_path / "example.meta.json"
+    passages_path = tmp_path / "example.passages.jsonl"
+    index_path = tmp_path / "example.passages.idx"
+
+    passages_path.write_text("first\n", encoding="utf-8")
+    index_path.write_bytes(b"index")
+    _write_meta(meta_path, passages_path.name, index_path.name, total=1)
+
+    # Initial start populates signature
+    ok, port = manager.start_server(
+        port=6000,
+        model_name="test-model",
+        passages_file=str(meta_path),
+    )
+    assert ok
+    assert port == 6000
+    assert len(start_calls) == 1
+
+    initial_signature = start_calls[0]["passages_signature"]
+
+    # No metadata change => reuse existing server
+    ok, port_again = manager.start_server(
+        port=6000,
+        model_name="test-model",
+        passages_file=str(meta_path),
+    )
+    assert ok
+    assert port_again == 6000
+    assert len(start_calls) == 1
+
+    # Modify passage data and metadata to force signature change
+    time.sleep(0.01)  # Ensure filesystem timestamps move forward
+    passages_path.write_text("second\n", encoding="utf-8")
+    _write_meta(meta_path, passages_path.name, index_path.name, total=2)
+
+    ok, port_third = manager.start_server(
+        port=6000,
+        model_name="test-model",
+        passages_file=str(meta_path),
+    )
+    assert ok
+    assert port_third == 6000
+    assert len(start_calls) == 2
+
+    updated_signature = start_calls[1]["passages_signature"]
+    assert updated_signature != initial_signature