Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Supplement for Aligner #261

Merged
merged 6 commits into from
Mar 24, 2024
Merged

Supplement for Aligner #261

Show file tree
Hide file tree
Changes from 5 commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
ac25e87
update benchmark results
AlignInc Mar 22, 2024
f2e43fe
pin cohere version
AlignInc Mar 22, 2024
28e6411
remove line in requirements.txt
AlignInc Mar 22, 2024
e2dabd8
add outputs and update leaderboard
AlignInc Mar 22, 2024
bdfad7b
fix minor name error
AlignInc Mar 22, 2024
74aea9d
remove unnecessary files
AlignInc Mar 24, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 7 additions & 0 deletions docs/data_AlpacaEval_2/claude_3_opus_ranking_leaderboard.csv
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,7 @@
name,length_controlled_winrate,win_rate,avg_length,link,samples,filter
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

no need to add docs/* file

GPT-4 Preview,50.0,50.0,2049,,https://github.com/tatsu-lab/alpaca_eval/blob/main/results/gpt4_1106_preview/model_outputs.json,community
Claude 3 Opus (02/29),43.25056335573304,27.45341614906832,1388,,https://github.com/tatsu-lab/alpaca_eval/blob/main/results/claude-3-opus-20240229/model_outputs.json,minimal
GPT-4 0314,29.779791079392187,15.217391304347828,1371,,https://github.com/tatsu-lab/alpaca_eval/blob/main/results/gpt4_0314/model_outputs.json,verified
Mistral Large (24/02),28.18279361879813,16.459627329192546,1362,https://mistral.ai/news/la-plateforme/,https://github.com/tatsu-lab/alpaca_eval/blob/main/results/mistral-large-2402/model_outputs.json,minimal
GPT-4 0613,20.456928802947065,8.136645962732919,1140,,https://github.com/tatsu-lab/alpaca_eval/blob/main/results/gpt4_0613/model_outputs.json,minimal
GPT 3.5 Turbo 1106,16.7339348632326,6.211180124223603,796,,https://github.com/tatsu-lab/alpaca_eval/blob/main/results/gpt-3.5-turbo-1106/model_outputs.json,minimal
7 changes: 7 additions & 0 deletions docs/data_AlpacaEval_2/mistral-large-2402_ranking_leaderboard.csv
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,7 @@
name,length_controlled_winrate,win_rate,avg_length,link,samples,filter
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

no need to add docs/* file

GPT-4 Preview,50.0,50.0,2049,,https://github.com/tatsu-lab/alpaca_eval/blob/main/results/gpt4_1106_preview/model_outputs.json,minimal
Claude 3 Opus (02/29),47.450744462524334,32.94723294723295,1388,,https://github.com/tatsu-lab/alpaca_eval/blob/main/results/claude-3-opus-20240229/model_outputs.json,minimal
Mistral Large (24/02),45.47989179343149,28.045515394912982,1362,https://mistral.ai/news/la-plateforme/,https://github.com/tatsu-lab/alpaca_eval/blob/main/results/mistral-large-2402/model_outputs.json,minimal
GPT-4 0314,40.88989260514661,25.32383419689119,1371,,https://github.com/tatsu-lab/alpaca_eval/blob/main/results/gpt4_0314/model_outputs.json,minimal
GPT-4 0613,34.303198292783584,18.14044213263979,1140,,https://github.com/tatsu-lab/alpaca_eval/blob/main/results/gpt4_0613/model_outputs.json,verified
GPT 3.5 Turbo 1106,28.929334810369056,11.558441558441558,796,,https://github.com/tatsu-lab/alpaca_eval/blob/main/results/gpt-3.5-turbo-1106/model_outputs.json,minimal
2 changes: 2 additions & 0 deletions docs/data_AlpacaEval_2/weighted_alpaca_eval_gpt4_turbo_leaderboard.csv
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,7 +1,9 @@
name,length_controlled_winrate,win_rate,avg_length,link,samples,filter
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

no need to add docs/* file

GPT-4 Preview,50.0,50.0,2049,,https://github.com/tatsu-lab/alpaca_eval/blob/main/results/gpt4_1106_preview/model_outputs.json,minimal
Aligner 2B+Claude 3 Opus,41.823071715247664,34.46337362321739,1669,https://github.com/AlignInc/aligner-replication,https://github.com/tatsu-lab/alpaca_eval/blob/main/results/aligner-2b_claude-3-opus-20240229/model_outputs.json,community
Claude 3 Opus (02/29),40.39177606350116,29.04176413403727,1388,,https://github.com/tatsu-lab/alpaca_eval/blob/main/results/claude-3-opus-20240229/model_outputs.json,minimal
GPT-4,38.12808974440021,23.576789314782605,1365,,https://github.com/tatsu-lab/alpaca_eval/blob/main/results/gpt4/model_outputs.json,minimal
Aligner 2B+Qwen1.5 72B Chat,36.725868878524274,31.773037737123104,1812,https://github.com/AlignInc/aligner-replication,https://github.com/tatsu-lab/alpaca_eval/blob/main/results/aligner-2b_qwen1.5-72b-chat/model_outputs.json,community
Qwen1.5 72B Chat,36.571754111987296,26.49828339562733,1549,https://huggingface.co/Qwen/Qwen1.5-72B-Chat,https://github.com/tatsu-lab/alpaca_eval/blob/main/results/Qwen1.5-72B-Chat/model_outputs.json,community
GPT-4 0314,35.30706121640206,22.073258928708075,1371,,https://github.com/tatsu-lab/alpaca_eval/blob/main/results/gpt4_0314/model_outputs.json,verified
Claude 3 Sonnet (02/29),34.87247436243302,25.556325292273296,1420,,https://github.com/tatsu-lab/alpaca_eval/blob/main/results/claude-3-sonnet-20240229/model_outputs.json,verified
Expand Down
4,832 changes: 4,832 additions & 0 deletions results/aligner-2b_claude-3-opus-20240229/model_outputs.json
View file
Open in desktop

Large diffs are not rendered by default.

58,452 changes: 58,452 additions & 0 deletions results/aligner-2b_claude-3-opus-20240229/weighted_alpaca_eval_gpt4_turbo/annotations.json
View file
Open in desktop

Large diffs are not rendered by default.

4,832 changes: 4,832 additions & 0 deletions results/aligner-2b_qwen1.5-72b-chat/model_outputs.json
View file
Open in desktop

Large diffs are not rendered by default.

55,281 changes: 55,281 additions & 0 deletions results/aligner-2b_qwen1.5-72b-chat/weighted_alpaca_eval_gpt4_turbo/annotations.json
View file
Open in desktop

Large diffs are not rendered by default.

1,502 changes: 1,502 additions & 0 deletions results_evaluators/tmp_test/tmp_test/annotation.json
View file
Open in desktop

Large diffs are not rendered by default.

31 changes: 31 additions & 0 deletions results_evaluators/tmp_test/tmp_test/leaderboard.csv
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,31 @@
,Human agreement,Price [$/1000 examples],Time [seconds/1000 examples],Spearman corr.,Pearson corr.,Bias,Variance,Proba. prefer longer,Proba. prefer lists,Proba. prefer 1,# parsed,mode
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

please remove this file

alpaca_eval_gpt4_fn,70.98765432098766,14.471944444444444,5046.056233900002,0.95,0.9447778041206524,27.623456790123456,11.11111111111111,0.750561797752809,0.6756756756756757,0.4799382716049383,2592,verified
improved_aviary_gpt4,69.75308641975309,12.781435185185186,1831.2850013,0.8833333333333333,0.8993690915590962,,,0.7280898876404495,0.7027027027027027,0.4861111111111111,648,verified
alpaca_eval_gpt4,69.1743827160494,13.601944444444444,1455.4169713998845,0.9666666666666668,0.9335485321531084,28.395061728395056,14.621913580246911,0.6831460674157304,0.7297297297297297,0.5011574074074074,2592,minimal
alpaca_eval_clf_cot_gpt4_turbo,68.70109546165884,6.441079812206572,1753.4788411931145,0.9333333333333332,0.7570054666164165,,,0.6863636363636364,0.6545454545454545,0.5352112676056338,639,verified
alpaca_eval_cot_gpt4_turbo_fn,68.63874533448178,6.311349574632637,1988.6012626717545,0.9707197941566388,0.8997919147215918,29.320987654320984,18.435272517819858,0.6696629213483146,0.6126126126126126,0.5232018561484919,2586,minimal
weighted_alpaca_eval_cot_gpt4_turbo,68.45771313115921,6.447465224111284,1869.2926495435856,0.9333333333333332,0.7743167748273401,,,0.6853932584269663,0.6576576576576577,0.5283575514995362,647,verified
aviary_gpt4,68.3641975308642,12.781481481481482,1821.0640311000004,0.9205101496312952,0.9053426857899228,,,0.701123595505618,0.6486486486486487,0.5555555555555556,648,verified
alpaca_eval_gpt4_turbo_fn,68.09413580246913,5.533981481481482,864.3023563021605,0.9333333333333332,0.817290435500228,30.246913580246915,15.625,0.651685393258427,0.6036036036036037,0.5381944444444444,2592,minimal
gpt4_turbo_cot_logprob,67.86974910317902,5.397145061728395,1568.9484159171295,0.6333333333333333,0.6310442120964042,,,0.5932584269662922,0.5855855855855856,0.5285319490509259,648,verified
gpt4_turbo_cot_clf,67.59689922480621,5.3972248062015495,1528.4046718706977,0.6666666666666667,0.6326057742256878,,,0.5936794582392777,0.5855855855855856,0.5255813953488373,645,verified
claude_ranking,67.5925925925926,4.954578395061729,218.4230414438272,0.9,0.90848221004591,,,0.7303370786516854,0.6576576576576577,0.4552469135802468,648,verified
gpt4,66.93672839506173,12.452592592592593,1036.788589334915,0.8833333333333333,0.8668599990267735,31.481481481481488,14.621913580246911,0.647191011235955,0.6666666666666666,0.5397376543209877,2592,minimal
alpaca_farm_greedy_gpt4,66.43518518518519,15.28163425925926,877.6250469425926,0.8499999999999999,0.7481465609199582,30.246913580246915,19.290123456790123,0.597752808988764,0.6486486486486487,0.5362654320987654,2592,minimal
weighted_alpaca_eval_gpt4_turbo,65.73198824263118,4.323981481481481,227.7462866895061,0.7833333333333333,0.7688872243700914,33.89896126543981,23.652705035108028,0.6058558558558559,0.5727272727272728,0.5282783420419752,2592,minimal
humans,65.66358024691358,300.0,36800.00000000001,1.0,1.0,0.0,34.336419753086425,0.6359550561797753,0.6036036036036037,0.5177469135802468,2592,minimal
gpt4_turbo_clf,65.58641975308642,3.774166666666667,157.86959398549385,0.5666666666666667,0.6056662735192052,,,0.5123595505617977,0.5405405405405406,0.5555555555555556,648,verified
alpaca_eval_clf_gpt4_turbo,65.42635658914729,4.328077519379845,151.46231159178296,0.7166666666666667,0.7351663293324147,,,0.6049661399548533,0.5909090909090909,0.5271317829457365,645,verified
claude,65.31635802469135,3.298695848765433,172.99865933897803,0.9333333333333332,0.9028603896845376,32.407407407407405,18.47993827160494,0.6606741573033708,0.6666666666666666,0.494212962962963,2592,minimal
lmsys_gpt4,65.25848765432099,13.945289351851851,17981.91908101215,0.9833333333333332,0.9656100250020464,31.59722222222222,15.91435185185185,0.7389277389277389,0.6944444444444444,0.4635416666666667,2592,minimal
gpt4_turbo,64.14219474497682,4.165919629057188,185.73029410061824,0.5666666666666667,0.5688213739495881,,,0.5382882882882883,0.5675675675675675,0.571870170015456,647,verified
text_davinci_003,64.0817901234568,8.712680555439814,120.90134619274691,0.8499999999999999,0.8307147459007311,33.796296296296305,22.72376543209876,0.6966292134831461,0.6576576576576577,0.4733796296296295,2592,minimal
gpt4_turbo_logprob,63.51076045576003,3.774166666666667,142.6550541719136,0.6166666666666666,0.6016102512172834,35.53043431362654,17.968493710574844,0.509009009009009,0.5225225225225225,0.5600111475683258,2592,verified
guanaco_33b,62.74944567627494,,910.8929739450112,0.0,0.2495312789260463,,,0.6991150442477876,0.704225352112676,0.4257206208425721,451,verified
improved_lmsys_gpt4,62.34567901234568,13.938055555555556,5397.837981725772,0.9833333333333332,0.9273862641854697,,,0.7534883720930232,0.7117117117117117,0.4490740740740742,648,verified
longest,62.19135802469136,0.0,0.0,0.2666666666666666,0.5604276915228803,37.808641975308646,0.0,1.0,0.8828828828828829,0.4166666666666667,2592,minimal
chatgpt_fn,59.992283950617285,1.0088333333333337,529.928419875,0.75,0.8270316070156506,36.88271604938272,27.739197530864203,0.6247191011235955,0.6216216216216216,0.4911265432098766,2592,verified
alpaca_farm,57.80525502318392,11.978385883565174,1312.895122694532,0.5272012675161055,0.6048080773927609,,,0.5900900900900901,0.5636363636363636,0.5100463678516229,647,verified
chatgpt,57.28201740503198,0.8342726921591347,284.9753823429895,0.7166666666666667,0.7136212819980075,39.35185185185186,34.054591087228026,0.5910112359550562,0.5945945945945946,0.488991888760139,2589,minimal
cohere,56.60964230171073,6.485108864696734,503.1591360234836,0.2166666666666666,0.4349894801752539,,,0.6281179138321995,0.6486486486486487,0.4603421461897357,643,verified
tmp_test,51.0,,,,,,,0.27941176470588236,0.3333333333333333,1.0,100,community
2 changes: 1 addition & 1 deletion setup.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -35,7 +35,7 @@
]
PACKAGES_ALL_API = [
"anthropic>=0.18",
"cohere",
"cohere<5.0.0a0",
"replicate",
"boto3>=1.28.58",
"google-generativeai",
Expand Down
2 changes: 2 additions & 0 deletions ...lpaca_eval/leaderboards/data_AlpacaEval_2/weighted_alpaca_eval_gpt4_turbo_leaderboard.csv
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,8 +2,10 @@
gpt4_1106_preview_verbose,64.30360147101865,1.3348590089025316,525,268,12,805,65.96273291925466,dev,2402,51.57500797967598
gpt4_1106_preview,50.0,0.0,0,0,805,805,50.0,minimal,2049,50.0
gpt4_1106_preview_concise,22.92019444047205,1.232517714329424,172,622,11,805,22.049689440993788,dev,1136,41.896601591245386
aligner-2b_claude-3-opus-20240229,34.46337362321739,1.314666526302454,225,475,105,805,34.47204968944099,community,1669,41.823071715247664
claude-3-opus-20240229,29.04176413403727,1.3942602231385623,223,580,2,805,27.82608695652174,minimal,1388,40.39177606350116
gpt4,23.576789314782605,1.275704201206918,179,618,8,805,22.732919254658384,minimal,1365,38.12808974440021
aligner-2b_qwen1.5-72b-chat,31.773037737123104,1.2392772646245978,180,473,152,805,31.801242236024844,community,1812,36.725868878524274
Qwen1.5-72B-Chat,26.49828339562733,1.304236164893057,201,600,4,805,25.217391304347824,community,1549,36.571754111987296
gpt4_0314,22.073258928708075,1.2466725494608204,172,627,6,805,21.73913043478261,verified,1371,35.30706121640206
claude-3-sonnet-20240229,25.556325292273296,1.3419811051815638,193,608,4,805,24.22360248447205,verified,1420,34.87247436243302
Expand Down
14 changes: 14 additions & 0 deletions src/alpaca_eval/models_configs/aligner-2b_claude-3-opus-20240229/configs.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,14 @@
aligner-2b_claude-3-opus-20240229:
prompt_template: "aligner-2b_claude-3-opus-20240229/prompt.txt"
fn_completions: null
completions_kwargs:
model_name: "./aligner_2b" # local path
model_kwargs:
torch_dtype: "bfloat16"
max_new_tokens: 8192
temperature: 0.05
top_p: 0.95
top_k: 10
do_sample: False
pretty_name: "Aligner 2B+Claude 3 Opus"
link: "https://github.com/AlignInc/aligner-replication"
1 change: 1 addition & 0 deletions src/alpaca_eval/models_configs/aligner-2b_claude-3-opus-20240229/prompt.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
BEGINNING OF CONVERSATION: USER: ##Question: {question} ##Answer: {answer} ##Revision: ASSISTANT:
14 changes: 14 additions & 0 deletions src/alpaca_eval/models_configs/aligner-2b_qwen1.5-72b-chat/configs.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,14 @@
aligner-2b_qwen1.5-72b-chat:
prompt_template: "aligner-2b_qwen1.5-72b-chat/prompt.txt"
fn_completions: null
completions_kwargs:
model_name: "./aligner_2b" # local path
model_kwargs:
torch_dtype: "bfloat16"
max_new_tokens: 8192
temperature: 0.05
top_p: 0.95
top_k: 10
do_sample: False
pretty_name: "Aligner 2B+Qwen1.5 72B Chat"
link: "https://github.com/AlignInc/aligner-replication"
1 change: 1 addition & 0 deletions src/alpaca_eval/models_configs/aligner-2b_qwen1.5-72b-chat/prompt.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
BEGINNING OF CONVERSATION: USER: ##Question: {question} ##Answer: {answer} ##Revision: ASSISTANT:
Loading