diff --git a/README.md b/README.md index 0f61021..03850f4 100644 --- a/README.md +++ b/README.md @@ -84,7 +84,7 @@ Additionally, we are recommending implementing a solution to mark the metadata o ## 🧨 Lower VRAM diffusers example -The below example should run on a RTX 4090. +The below example should run on a RTX 4090. For more examples check the [diffusers quantization guide here](docs/flux2_dev_hf.md) ```python import torch diff --git a/docs/flux2_dev_hf.md b/docs/flux2_dev_hf.md index 4c5ee75..a4f9861 100644 --- a/docs/flux2_dev_hf.md +++ b/docs/flux2_dev_hf.md @@ -8,7 +8,7 @@ Install diffusers from `main` pip install git+https://github.com/huggingface/diffusers.git ``` -After accepting the gating on this repository, login with Hugging Face on your terminal +After accepting the gating on the [FLUX.2-dev repository](https://huggingface.co/black-forest-labs/FLUX.2-dev), login with Hugging Face on your terminal ```sh hf auth login ``` @@ -28,13 +28,13 @@ The text-embeddings are calculated in bf16 in the cloud and you only load the tr ```py import torch -from diffusers import Flux2Pipeline, Flux2Transformer2DModel +from diffusers import Flux2Pipeline from diffusers.utils import load_image from huggingface_hub import get_token import requests import io -repo_id = "diffusers/FLUX.2-dev-bnb-4bit" +repo_id = "diffusers/FLUX.2-dev-bnb-4bit" #quantized text-encoder and DiT. VAE still in bf16 device = "cuda:0" torch_dtype = torch.bfloat16 @@ -52,14 +52,15 @@ def remote_text_encoder(prompts): return prompt_embeds.to(device) pipe = Flux2Pipeline.from_pretrained( - repo_id, transformer=transformer, text_encoder=None, torch_dtype=torch_dtype + repo_id, text_encoder=None, torch_dtype=torch_dtype ).to(device) prompt = "Realistic macro photograph of a hermit crab using a soda can as its shell, partially emerging from the can, captured with sharp detail and natural colors, on a sunlit beach with soft shadows and a shallow depth of field, with blurred ocean waves in the background. The can has the text `BFL Diffusers` on it and it has a color gradient that start with #FF5733 at the top and transitions to #33FF57 at the bottom." +#cat_image = load_image("https://huggingface.co/spaces/zerogpu-aoti/FLUX.1-Kontext-Dev-fp8-dynamic/resolve/main/cat.png") image = pipe( prompt_embeds=remote_text_encoder(prompt), - #image=load_image("https://huggingface.co/spaces/zerogpu-aoti/FLUX.1-Kontext-Dev-fp8-dynamic/resolve/main/cat.png") #optional image input + #image=load_image(cat_image) #optional image input generator=torch.Generator(device=device).manual_seed(42), num_inference_steps=50, #28 steps can be a good trade-off guidance_scale=4, @@ -75,30 +76,23 @@ The text-encoder is offloaded from VRAM for the transformer to run with `pipe.en ```py import torch -from transformers import Mistral3ForConditionalGeneration -from diffusers import Flux2Pipeline, Flux2Transformer2DModel +from diffusers import Flux2Pipeline +from diffusers.utils import load_image -repo_id = "diffusers/FLUX.2-dev-bnb-4bit" +repo_id = "diffusers/FLUX.2-dev-bnb-4bit" #quantized text-encoder and DiT. VAE still in bf16 device = "cuda:0" torch_dtype = torch.bfloat16 -transformer = Flux2Transformer2DModel.from_pretrained( - repo_id, subfolder="transformer", torch_dtype=torch_dtype, device_map="cpu" -) -text_encoder = Mistral3ForConditionalGeneration.from_pretrained( - repo_id, subfolder="text_encoder", dtype=torch_dtype, device_map="cpu" -) - pipe = Flux2Pipeline.from_pretrained( - repo_id, transformer=transformer, text_encoder=text_encoder, torch_dtype=torch_dtype + repo_id, torch_dtype=torch_dtype ) pipe.enable_model_cpu_offload() prompt = "Realistic macro photograph of a hermit crab using a soda can as its shell, partially emerging from the can, captured with sharp detail and natural colors, on a sunlit beach with soft shadows and a shallow depth of field, with blurred ocean waves in the background. The can has the text `BFL + Diffusers` on it and it has a color gradient that start with #FF5733 at the top and transitions to #33FF57 at the bottom." - +#cat_image = load_image("https://huggingface.co/spaces/zerogpu-aoti/FLUX.1-Kontext-Dev-fp8-dynamic/resolve/main/cat.png") image = pipe( prompt=prompt, - #image=[load_image("https://huggingface.co/spaces/zerogpu-aoti/FLUX.1-Kontext-Dev-fp8-dynamic/resolve/main/cat.png")] #multi-image input + #image=[cat_image] #multi-image input generator=torch.Generator(device=device).manual_seed(42), num_inference_steps=50, guidance_scale=4, @@ -113,12 +107,13 @@ To understand how different quantizations affect the model's abilities and quali ## 💿 More VRAM (80G+) -Even an H100 can't hold the text-encoder, transormer and VAE at the same time. However, here it is a matter of activating the `pipe.enable_model_cpu_offload()` -And for H200, B200 or larger carts, everything fits. +Even an H100 can't hold the text-encoder, transormer and VAE at the same time. However, as they each fit individually, it is a matter of activating the `pipe.enable_model_cpu_offload()` +For H200, B200 or larger cards, everything fits. ```py import torch from diffusers import Flux2Pipeline +from diffusers.utils import load_image repo_id = "black-forest-labs/FLUX.2-dev" device = "cuda:0" @@ -127,13 +122,14 @@ torch_dtype = torch.bfloat16 pipe = Flux2Pipeline.from_pretrained( repo_id, torch_dtype=torch_dtype ) -pipe.enable_model_cpu_offload() #deactivate for >80G VRAM carts like H200, B200, etc. and do a `pipe.to(device)` instead +pipe.enable_model_cpu_offload() #no need to do cpu offload for >80G VRAM carts like H200, B200, etc. and do a `pipe.to(device)` instead prompt = "Realistic macro photograph of a hermit crab using a soda can as its shell, partially emerging from the can, captured with sharp detail and natural colors, on a sunlit beach with soft shadows and a shallow depth of field, with blurred ocean waves in the background. The can has the text `BFL Diffusers` on it and it has a color gradient that start with #FF5733 at the top and transitions to #33FF57 at the bottom." +#cat_image = load_image("https://huggingface.co/spaces/zerogpu-aoti/FLUX.1-Kontext-Dev-fp8-dynamic/resolve/main/cat.png") image = pipe( prompt=prompt, - #image=[load_image("https://huggingface.co/spaces/zerogpu-aoti/FLUX.1-Kontext-Dev-fp8-dynamic/resolve/main/cat.png")] #multi-image input + #image=[cat_image] #multi-image input generator=torch.Generator(device=device).manual_seed(42), num_inference_steps=50, guidance_scale=4, @@ -146,7 +142,8 @@ image.save("flux2_output.png") `pipe.enable_model_cpu_offload()` slows you down a bit. You can move as fast as possible on the H100 with the remote text-encoder ```py import torch -from diffusers import Flux2Pipeline, Flux2Transformer2DModel +from diffusers import Flux2Pipeline +from diffusers.utils import load_image from huggingface_hub import get_token import requests import io @@ -175,9 +172,10 @@ pipe = Flux2Pipeline.from_pretrained( prompt = "Realistic macro photograph of a hermit crab using a soda can as its shell, partially emerging from the can, captured with sharp detail and natural colors, on a sunlit beach with soft shadows and a shallow depth of field, with blurred ocean waves in the background. The can has the text `BFL + Diffusers` on it and it has a color gradient that start with #FF5733 at the top and transitions to #33FF57 at the bottom." +#cat_image = load_image("https://huggingface.co/spaces/zerogpu-aoti/FLUX.1-Kontext-Dev-fp8-dynamic/resolve/main/cat.png") image = pipe( prompt_embeds=remote_text_encoder(prompt), - #image=[load_image("https://huggingface.co/spaces/zerogpu-aoti/FLUX.1-Kontext-Dev-fp8-dynamic/resolve/main/cat.png")] #optional multi-image input + #image=[cat_image] #optional multi-image input generator=torch.Generator(device=device).manual_seed(42), num_inference_steps=50, guidance_scale=4, @@ -188,4 +186,4 @@ image.save("flux2_output.png") ## 🧮 Other VRAM sizes -If you have different GPU sizes, you can experiment with different quantizations, for example, for 40-48G VRAM GPUs, (8-bit) quantization instead of 4-bit can be a good trade-off. You can learn more on the [diffusers FLUX.2 release blog](https://huggingface.co/blog/flux2) \ No newline at end of file +If you have different GPU sizes, you can experiment with different quantizations, for example, for 40-48G VRAM GPUs, (8-bit) quantization instead of 4-bit can be a good trade-off. You can learn more on the [diffusers FLUX.2 release blog](https://huggingface.co/blog/flux2)