Initial commit

This commit is contained in:
yichuan520030910320
2025-06-30 09:05:05 +00:00
commit 46f6cc100b
1231 changed files with 278432 additions and 0 deletions

View File

@@ -0,0 +1,4 @@
Model,NQ Exact Match,NQ F1,TriviaQA Exact Match,TriviaQA F1,GPQA Exact Match,GPQA F1,HotpotQA Exact Match,HotpotQA F1
BM25,0.192,0.277,0.406,0.474,0.020089,0.04524,0.162,0.239
PQ 5,0.2075,0.291,0.422,0.495,0.0201,0.0445,0.148,0.219
Ours,0.265,0.361,0.533,0.604,0.02008,0.0452,0.182,0.2729
1 Model NQ Exact Match NQ F1 TriviaQA Exact Match TriviaQA F1 GPQA Exact Match GPQA F1 HotpotQA Exact Match HotpotQA F1
2 BM25 0.192 0.277 0.406 0.474 0.020089 0.04524 0.162 0.239
3 PQ 5 0.2075 0.291 0.422 0.495 0.0201 0.0445 0.148 0.219
4 Ours 0.265 0.361 0.533 0.604 0.02008 0.0452 0.182 0.2729

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:1296720e79196bbdf38f051043c1b054667803726a24036c0b6a87cedb204ea5
size 227482438

View File

@@ -0,0 +1,21 @@
2,1,512,1024,0.541,0.326,1.659509202
2,2,512,1024,0.979,0.621,1.576489533
2,4,512,1024,1.846,0.977,1.889457523
2,8,512,1024,3.575,1.943,1.83993824
2,16,512,1024,7.035,3.733,1.884543263
2,32,512,1024,15.655,8.517,1.838088529
2,64,512,1024,32.772,17.43,1.88020654
4,1,512,1024,2.675,1.38,1.938405797
4,2,512,1024,5.397,2.339,2.307396323
4,4,512,1024,10.672,4.944,2.158576052
4,8,512,1024,21.061,9.266,2.272933305
4,16,512,1024,46.332,18.334,2.527108105
4,32,512,1024,99.607,36.156,2.754923111
4,64,512,1024,186.348,72.356,2.575432583
8,1,512,1024,7.325,4.087,1.792268167
8,2,512,1024,14.109,7.491,1.883460152
8,4,512,1024,28.499,14.013,2.033754371
8,8,512,1024,65.222,27.453,2.375769497
8,16,512,1024,146.294,52.55,2.783901047
8,32,512,1024,277.099,103.61,2.674442621
8,64,512,1024,512.979,208.36,2.461984066
1 2 1 512 1024 0.541 0.326 1.659509202
2 2 2 512 1024 0.979 0.621 1.576489533
3 2 4 512 1024 1.846 0.977 1.889457523
4 2 8 512 1024 3.575 1.943 1.83993824
5 2 16 512 1024 7.035 3.733 1.884543263
6 2 32 512 1024 15.655 8.517 1.838088529
7 2 64 512 1024 32.772 17.43 1.88020654
8 4 1 512 1024 2.675 1.38 1.938405797
9 4 2 512 1024 5.397 2.339 2.307396323
10 4 4 512 1024 10.672 4.944 2.158576052
11 4 8 512 1024 21.061 9.266 2.272933305
12 4 16 512 1024 46.332 18.334 2.527108105
13 4 32 512 1024 99.607 36.156 2.754923111
14 4 64 512 1024 186.348 72.356 2.575432583
15 8 1 512 1024 7.325 4.087 1.792268167
16 8 2 512 1024 14.109 7.491 1.883460152
17 8 4 512 1024 28.499 14.013 2.033754371
18 8 8 512 1024 65.222 27.453 2.375769497
19 8 16 512 1024 146.294 52.55 2.783901047
20 8 32 512 1024 277.099 103.61 2.674442621
21 8 64 512 1024 512.979 208.36 2.461984066

View File

@@ -0,0 +1,9 @@
Dataset,Metric,Original,original + batch,original + two_level,original + two_level + batch
NQ,Latency,6.9,5.8,4.2,3.7
NQ,SpeedUp,1,1.18965517,1.64285714,1.86486486
TriviaQA,Latency,17.054,14.542,12.046,10.83
TriviaQA,SpeedUp,1,1.17274103,1.41573967,1.57469990
GPQA,Latency,9.164,7.639,6.798,5.77
GPQA,SpeedUp,1,1.19963346,1.34804354,1.58821490
HotpotQA,Latency,60.279,39.827,50.664,29.868
HotpotQA,SpeedUp,1,1.51352098,1.18977972,2.01817999
1 Dataset Metric Original original + batch original + two_level original + two_level + batch
2 NQ Latency 6.9 5.8 4.2 3.7
3 NQ SpeedUp 1 1.18965517 1.64285714 1.86486486
4 TriviaQA Latency 17.054 14.542 12.046 10.83
5 TriviaQA SpeedUp 1 1.17274103 1.41573967 1.57469990
6 GPQA Latency 9.164 7.639 6.798 5.77
7 GPQA SpeedUp 1 1.19963346 1.34804354 1.58821490
8 HotpotQA Latency 60.279 39.827 50.664 29.868
9 HotpotQA SpeedUp 1 1.51352098 1.18977972 2.01817999

View File

@@ -0,0 +1,25 @@
Dataset,Hardware,Recall_target,HNSW,IVF,DiskANN,IVF-Disk,IVF-Recompute,Our,BM25,LLM_Gen_Time_1B,LLM_Gen_Time_3B,LLM_Gen_Time_7B
NQ,A10,85%,0.046,1.656,0.017,2.996,482.53,3.323,0.021,0.085,0.217,0.472
NQ,A10,90%,0.051,2.552,0.028,3.437,769.04,4.616,0,0.085,0.217,0.472
NQ,A10,95%,0.055,5.163,0.070,5.602,1436.26,19.494,0,0.085,0.217,0.472
NQ,MAC,85%,0,0,0.152,2.199,1535.10,7.971,0.033,0.316,0.717,1.468
NQ,MAC,90%,0,0,0.37,2.936,2446.60,13.843,0,0.316,0.717,1.468
NQ,MAC,95%,0,0,1.207,4.191,4569.29,44.363,0,0.316,0.717,1.468
TriviaQA,A10,85%,0.042,1.772,0.032,2.464,560.5,3.752,0.033,0.139,0.156,0.315
TriviaQA,A10,90%,0.043,3.541,0.057,3.651,997.81,5.777,0,0.139,0.156,0.315
TriviaQA,A10,95%,0.053,7.168,0.090,5.458,2005.33,20.944,0,0.139,0.156,0.315
TriviaQA,MAC,85%,0,0,0.481,1.875,1783.14787,8.889,0.036,0.325,0.692,1.415
TriviaQA,MAC,90%,0,0,0.984,2.639,3174.410301,17.145,0,0.325,0.692,1.415
TriviaQA,MAC,95%,0,0,1.578,3.884,6379.712245,47.909,0,0.325,0.692,1.415
GPQA,A10,85%,0.041,0.134,0.024,0.048,40.16,1.897,0.137,0.443,0.396,0.651
GPQA,A10,90%,0.042,0.174,0.034,0.06,54.71,1.733,0,0.443,0.396,0.651
GPQA,A10,95%,0.045,0.292,0.051,0.11,97.67,4.033,0,0.443,0.396,0.651
GPQA,MAC,85%,0,0,0.144,0.087,127.7707505,4.762,0.100,0.37,0.813,1.676
GPQA,MAC,90%,0,0,0.288,0.108,174.0647409,5.223,0,0.37,0.813,1.676
GPQA,MAC,95%,0,0,0.497,0.132,310.7380142,9.715,0,0.37,0.813,1.676
HotpotQA,A10,85%,0.044,2.519,0.054,4.048,724.26,10.358,0.70,0.144,0.196,0.420
HotpotQA,A10,90%,0.049,3.867,0.109,5.045,1173.67,15.515,0,0.144,0.196,0.420
HotpotQA,A10,95%,0.07,10.928,0.412,8.659,3079.57,61.757,0,0.144,0.196,0.420
HotpotQA,MAC,85%,0,0,0.974,2.844,2304.125187,23.636,0.052,0.144,0.196,0.420
HotpotQA,MAC,90%,0,0,1.913,3.542,3415.736201,44.803,0,0.144,0.196,0.420
HotpotQA,MAC,95%,0,0,5.783,6.764,9797.244043,140.62,0,0.144,0.196,0.420
1 Dataset Hardware Recall_target HNSW IVF DiskANN IVF-Disk IVF-Recompute Our BM25 LLM_Gen_Time_1B LLM_Gen_Time_3B LLM_Gen_Time_7B
2 NQ A10 85% 0.046 1.656 0.017 2.996 482.53 3.323 0.021 0.085 0.217 0.472
3 NQ A10 90% 0.051 2.552 0.028 3.437 769.04 4.616 0 0.085 0.217 0.472
4 NQ A10 95% 0.055 5.163 0.070 5.602 1436.26 19.494 0 0.085 0.217 0.472
5 NQ MAC 85% 0 0 0.152 2.199 1535.10 7.971 0.033 0.316 0.717 1.468
6 NQ MAC 90% 0 0 0.37 2.936 2446.60 13.843 0 0.316 0.717 1.468
7 NQ MAC 95% 0 0 1.207 4.191 4569.29 44.363 0 0.316 0.717 1.468
8 TriviaQA A10 85% 0.042 1.772 0.032 2.464 560.5 3.752 0.033 0.139 0.156 0.315
9 TriviaQA A10 90% 0.043 3.541 0.057 3.651 997.81 5.777 0 0.139 0.156 0.315
10 TriviaQA A10 95% 0.053 7.168 0.090 5.458 2005.33 20.944 0 0.139 0.156 0.315
11 TriviaQA MAC 85% 0 0 0.481 1.875 1783.14787 8.889 0.036 0.325 0.692 1.415
12 TriviaQA MAC 90% 0 0 0.984 2.639 3174.410301 17.145 0 0.325 0.692 1.415
13 TriviaQA MAC 95% 0 0 1.578 3.884 6379.712245 47.909 0 0.325 0.692 1.415
14 GPQA A10 85% 0.041 0.134 0.024 0.048 40.16 1.897 0.137 0.443 0.396 0.651
15 GPQA A10 90% 0.042 0.174 0.034 0.06 54.71 1.733 0 0.443 0.396 0.651
16 GPQA A10 95% 0.045 0.292 0.051 0.11 97.67 4.033 0 0.443 0.396 0.651
17 GPQA MAC 85% 0 0 0.144 0.087 127.7707505 4.762 0.100 0.37 0.813 1.676
18 GPQA MAC 90% 0 0 0.288 0.108 174.0647409 5.223 0 0.37 0.813 1.676
19 GPQA MAC 95% 0 0 0.497 0.132 310.7380142 9.715 0 0.37 0.813 1.676
20 HotpotQA A10 85% 0.044 2.519 0.054 4.048 724.26 10.358 0.70 0.144 0.196 0.420
21 HotpotQA A10 90% 0.049 3.867 0.109 5.045 1173.67 15.515 0 0.144 0.196 0.420
22 HotpotQA A10 95% 0.07 10.928 0.412 8.659 3079.57 61.757 0 0.144 0.196 0.420
23 HotpotQA MAC 85% 0 0 0.974 2.844 2304.125187 23.636 0.052 0.144 0.196 0.420
24 HotpotQA MAC 90% 0 0 1.913 3.542 3415.736201 44.803 0 0.144 0.196 0.420
25 HotpotQA MAC 95% 0 0 5.783 6.764 9797.244043 140.62 0 0.144 0.196 0.420

View File

@@ -0,0 +1,25 @@
Dataset,Hardware,Recall_target,HNSW,IVF,DiskANN,IVF-Disk,IVF-Recompute,Our,
NQ,A10,85%,0.046,1.656,0.017,2.996,482.53,4.243,
NQ,A10,90%,0.051,2.552,0.028,3.437,769.04,8.136,
NQ,A10,95%,0.055,5.163,0.070,5.602,1436.26,27.275,
NQ,MAC,85%,0,0,0.152,2.199,1535.10,10.672,
NQ,MAC,90%,0,0,0.37,2.936,2446.60,19.941,
NQ,MAC,95%,0,0,1.207,4.191,4569.29,61.383,
TriviaQA,A10,85%,0.042,1.772,0.032,2.464,560.5,5.612,
TriviaQA,A10,90%,0.043,3.541,0.057,3.651,997.81,10.737,
TriviaQA,A10,95%,0.053,7.168,0.090,5.458,2005.33,36.387,
TriviaQA,MAC,85%,0,0,0.481,1.875,1783.14787,12.825,
TriviaQA,MAC,90%,0,0,0.984,2.639,3174.410301,24.977,
TriviaQA,MAC,95%,0,0,1.578,3.884,6379.712245,85.734,
GPQA,A10,85%,0.041,0.134,0.024,0.048,40.16,2.269,
GPQA,A10,90%,0.042,0.174,0.034,0.06,54.71,3.200,
GPQA,A10,95%,0.045,0.292,0.051,0.11,97.67,7.445,
GPQA,MAC,85%,0,0,0.144,0.087,127.7707505,6.123,
GPQA,MAC,90%,0,0,0.288,0.108,174.0647409,8.507,
GPQA,MAC,95%,0,0,0.497,0.132,310.7380142,19.577,
HotpotQA,A10,85%,0.044,2.519,0.054,4.048,724.26,14.713,
HotpotQA,A10,90%,0.049,3.867,0.109,5.045,1173.67,33.561,
HotpotQA,A10,95%,0.07,10.928,0.412,8.659,3079.57,68.626,
HotpotQA,MAC,85%,0,0,0.974,2.844,2304.125187,34.783,
HotpotQA,MAC,90%,0,0,1.913,3.542,3415.736201,53.004,
HotpotQA,MAC,95%,0,0,5.783,6.764,9797.244043,95.413,
1 Dataset Hardware Recall_target HNSW IVF DiskANN IVF-Disk IVF-Recompute Our
2 NQ A10 85% 0.046 1.656 0.017 2.996 482.53 4.243
3 NQ A10 90% 0.051 2.552 0.028 3.437 769.04 8.136
4 NQ A10 95% 0.055 5.163 0.070 5.602 1436.26 27.275
5 NQ MAC 85% 0 0 0.152 2.199 1535.10 10.672
6 NQ MAC 90% 0 0 0.37 2.936 2446.60 19.941
7 NQ MAC 95% 0 0 1.207 4.191 4569.29 61.383
8 TriviaQA A10 85% 0.042 1.772 0.032 2.464 560.5 5.612
9 TriviaQA A10 90% 0.043 3.541 0.057 3.651 997.81 10.737
10 TriviaQA A10 95% 0.053 7.168 0.090 5.458 2005.33 36.387
11 TriviaQA MAC 85% 0 0 0.481 1.875 1783.14787 12.825
12 TriviaQA MAC 90% 0 0 0.984 2.639 3174.410301 24.977
13 TriviaQA MAC 95% 0 0 1.578 3.884 6379.712245 85.734
14 GPQA A10 85% 0.041 0.134 0.024 0.048 40.16 2.269
15 GPQA A10 90% 0.042 0.174 0.034 0.06 54.71 3.200
16 GPQA A10 95% 0.045 0.292 0.051 0.11 97.67 7.445
17 GPQA MAC 85% 0 0 0.144 0.087 127.7707505 6.123
18 GPQA MAC 90% 0 0 0.288 0.108 174.0647409 8.507
19 GPQA MAC 95% 0 0 0.497 0.132 310.7380142 19.577
20 HotpotQA A10 85% 0.044 2.519 0.054 4.048 724.26 14.713
21 HotpotQA A10 90% 0.049 3.867 0.109 5.045 1173.67 33.561
22 HotpotQA A10 95% 0.07 10.928 0.412 8.659 3079.57 68.626
23 HotpotQA MAC 85% 0 0 0.974 2.844 2304.125187 34.783
24 HotpotQA MAC 90% 0 0 1.913 3.542 3415.736201 53.004
25 HotpotQA MAC 95% 0 0 5.783 6.764 9797.244043 95.413

View File

@@ -0,0 +1,3 @@
Hardware,HNSW,IVF,DiskANN,IVF-Disk,IVF-Recompute,Our,BM25
RAM,190,171,10,0,0,0,0
Storage,185.4,171,240,171,0.5,5,59
1 Hardware HNSW IVF DiskANN IVF-Disk IVF-Recompute Our BM25
2 RAM 190 171 10 0 0 0 0
3 Storage 185.4 171 240 171 0.5 5 59

View File

@@ -0,0 +1,12 @@
Torch,8,55.592
Torch,16,75.439
Torch,32,110.025
Torch,64,186.496
Tutel,8,56.718
Tutel,16,82.121
Tutel,32,125.070
Tutel,64,216.191
BRT,8,56.725
BRT,16,79.291
BRT,32,93.180
BRT,64,118.923
1 Torch 8 55.592
2 Torch 16 75.439
3 Torch 32 110.025
4 Torch 64 186.496
5 Tutel 8 56.718
6 Tutel 16 82.121
7 Tutel 32 125.070
8 Tutel 64 216.191
9 BRT 8 56.725
10 BRT 16 79.291
11 BRT 32 93.180
12 BRT 64 118.923

View File

@@ -0,0 +1,6 @@
Disk cache size,0,2.5%(180G*2.5%),5%,8%,10%
Latency,,,,,
NQ,4.616,4.133,3.826,3.511,3.323
TriviaQA,5.777,4.979,4.553,4.141,3.916
GPQA,1.733,1.593,1.468,1.336,1.259
Hotpot,15.515,13.479,12.383,11.216,10.606
1 Disk cache size 0 2.5%(180G*2.5%) 5% 8% 10%
2 Latency
3 NQ 4.616 4.133 3.826 3.511 3.323
4 TriviaQA 5.777 4.979 4.553 4.141 3.916
5 GPQA 1.733 1.593 1.468 1.336 1.259
6 Hotpot 15.515 13.479 12.383 11.216 10.606