Add ExPO results to AlpacaEval

src/alpaca_eval/leaderboards/data_AlpacaEval_2/weighted_alpaca_eval_gpt4_turbo_leaderboard.csv

@@ -19,6 +19,7 @@ FsfairX-Zephyr-Chat-v0.1,35.94648644102434,1.4410058098036145,285,517,3,805,35.5
 Meta-Llama-3-70B-Instruct,33.17785695886864,1.3886514096065603,266,537,2,805,33.16770186335404,minimal,1919,34.42459717459881
 gpt4_0613_verbose,23.237360043453418,1.283539505582624,171,630,4,805,21.490683229813666,dev,1473,33.82126688658535
 mistral-large-2402,21.43877598137888,1.2485232545097724,166,638,1,805,20.6832298136646,verified,1362,32.65207998531868
+SPPO-Mistral7B-PairRM-ExPO,35.4431306716895,1.398130896602677,274,531,0,805,34.037267080745345,community,2288,31.822321960655582
 Samba-CoE-v0.2-best-of-16,26.988254318335404,1.3189030000371738,201,601,3,805,25.15527950310559,community,1578,31.506544268148147
 Mixtral-8x22B-Instruct-v0.1,22.21017054750302,1.2780740057417268,174,628,3,805,21.801242236024844,verified,1445,30.878810294279383
 SPPO-Mistral7B-PairRM,32.2453123637764,1.3908000109577154,249,556,0,805,30.93167701863354,community,2114,30.494137965217426
@@ -31,8 +32,11 @@ mistral-medium,21.855772543652176,1.2682402187223842,164,639,2,805,20.4968944099
 claude-2,17.188240356708075,1.17482825615589,131,673,1,805,16.335403726708076,verified,1069,28.155196141629148
 Samba-CoE-v0.2,21.847378669267083,1.2171089783436106,159,645,1,805,19.81366459627329,community,1469,27.62426735006872
 claude,16.98534361236025,1.1687959793014906,129,676,0,805,16.024844720496894,verified,1082,27.289504443727107
+internlm2-chat-20b-ExPO,46.185367468861,1.4638315245977938,375,430,0,805,46.58385093167702,community,3335,27.225759480731792
 Yi-34B-Chat,29.65994671879504,1.3225712597906096,219,582,4,805,27.45341614906832,verified,2123,27.19054787762733
+Starling-LM-7B-beta-ExPO,29.600851847906423,1.3252049542916096,225,580,0,805,27.95031055900621,community,2215,26.411156713811028
 Snorkel-Mistral-PairRM-DPO,30.220052700671644,1.3328273012530358,231,572,1,804,28.79353233830846,community,2736,26.39144645733206
+tulu-2-dpo-70b-ExPO,22.980619706104978,1.3591734082562228,184,620,1,805,22.919254658385093,community,1738,25.72330817134933
 claude-instant-1.2,16.12739962159006,1.1341036838301686,120,682,3,805,15.093167701863356,community,1112,25.61225902543337
 dbrx-instruct,18.44834898407453,1.255388020324377,150,655,0,805,18.633540372670808,verified,1450,25.37544974044448
 claude-2.1,15.733506736409938,1.120315865445773,115,688,2,805,14.409937888198757,verified,1096,25.251943886133027
@@ -47,12 +51,14 @@ Mixtral-8x7B-Instruct-v0.1_concise,13.744040154795034,1.071868299237546,105,700,
 Meta-Llama-3-8B-Instruct,22.56990260938061,1.257580233106669,176,626,3,805,22.049689440993788,minimal,1899,22.918784673210016
 Samba-CoE-v0.1,16.835501870062114,1.1180386124646702,124,680,1,805,15.46583850931677,community,1316,22.865837334795227
 gpt-3.5-turbo-16k-0613,14.13239070746584,1.027579400264853,96,704,5,805,12.236024844720497,verified,1328,22.720189163383225
+internlm2-chat-7b-ExPO,28.067817437082898,1.3159792318125112,209,595,1,805,26.02484472049689,community,2390,22.66748024879648
 gpt-3.5-turbo-0613,14.09579857390062,1.0371186215049395,99,700,6,805,12.670807453416147,community,1331,22.35251298054288
 gpt-3.5-turbo-1106_verbose,12.76316981026087,1.044246819212278,94,709,2,805,11.801242236024844,dev,1058,22.00093702171442
 gpt4_0613_concise,9.400320574596272,0.901021275896262,71,729,5,805,9.130434782608695,dev,627,21.57799091454269
 pairrm-tulu-2-70b,18.638962967441,1.1924966700012911,140,665,0,805,17.391304347826086,community,1607,21.428403975507223
 tulu-2-dpo-70b,15.982854374136648,1.1457861368237434,119,683,3,805,14.96894409937888,verified,1418,21.238610038371124
 Mistral-7B-ReMax-v0.1,15.999331369031056,1.1288683901451453,120,683,2,805,15.031055900621118,community,1478,20.55136770233589
+Starling-LM-7B-alpha-ExPO,18.179755920362158,1.2498324795896385,148,657,0,805,18.385093167701864,community,1821,19.4741654606294
 gpt-3.5-turbo-1106,9.177964561962735,0.8904117511864436,64,737,4,805,8.198757763975156,verified,796,19.30058903498905
 LMCocktail-10.7B-v1,13.153430917391304,1.045719535661201,104,700,1,805,12.981366459627331,community,1203,18.950710386651053
 internlm2-chat-20b-ppo,21.74915450048448,1.2443662409548863,170,632,3,805,21.30434782608696,community,2373,18.748739485433603
@@ -61,6 +67,7 @@ gpt-3.5-turbo-0301,9.622453295105588,0.9129656686751644,71,733,1,805,8.881987577
 xwinlm-13b-v0.1,17.42793475019876,1.1450161466942668,129,672,4,805,16.273291925465838,community,1894,17.918937898189796
 deepseek-llm-67b-chat,12.093422264919258,1.017384363293138,90,713,2,805,11.304347826086955,community,1151,17.843384089909343
 gpt35_turbo_instruct,8.462446504415423,0.8724086933609648,66,735,3,804,8.395522388059701,community,1018,17.72780108286588
+tulu-2-dpo-13b-ExPO,15.551405429399557,1.171485338425437,121,679,5,805,15.341614906832298,community,1649,17.591404469940848
 wizardlm-70b,14.383896086782608,1.0395048912985754,106,697,2,805,13.291925465838508,community,1545,17.575060737493747
 vicuna-33b-v1.3,12.705947921540371,0.999255784310268,90,711,4,805,11.428571428571429,verified,1479,17.574575310874923
 pairrm-tulu-2-13b,13.831901016757762,1.0835284665170843,110,694,1,805,13.72670807453416,community,1454,17.40520369795085
@@ -82,7 +89,9 @@ llama-2-70b-chat-hf,13.88825834374378,1.079984772728814,104,700,0,804,12.9353233
 openchat-v3.1-13b,11.082230489416148,0.9501308701291292,80,720,5,805,10.248447204968944,community,1484,14.50338795683784
 wizardlm-13b-v1.2,12.027480342770186,0.971761817748135,82,720,3,805,10.372670807453416,community,1635,14.462590694316631
 ultralm-13b-v2.0-best-of-16,13.853373471242236,1.049344706038026,98,705,2,805,12.298136645962732,community,1720,14.198987566645036
+zephyr-7b-beta-ExPO,11.06111683239833,1.0204784889272769,89,716,0,805,11.055900621118013,community,1405,14.001211980232686
 wizardlm-13b-v1.1,11.233909572857142,0.95027112458742,79,723,3,805,10.0,community,1525,13.91572059284851
+zephyr-7b-alpha-ExPO,10.55935434569986,0.977463444873356,79,725,1,805,9.875776397515528,community,1248,13.573089356781388
 zephyr-7b-beta,10.992885755354038,0.9617876718039866,78,725,2,805,9.813664596273291,community,1444,13.203198493136666
 dolphin-2.2.1-mistral-7b,9.039799728223604,0.8892901246776709,68,734,3,805,8.633540372670808,community,1130,13.121477650433736
 humpback-llama-65b,9.425139047801242,0.9300866722901956,70,734,1,805,8.75776397515528,community,1232,12.799859995893623
@@ -92,6 +101,7 @@ Qwen-14B-Chat,7.502333484720497,0.8147265702205473,57,742,6,805,7.45341614906832
 gpt4_gamed,3.7383373713788814,0.6278799633668313,32,771,2,805,4.099378881987578,community,68,12.188764057640531
 cut-13b,10.779089202496897,0.9428953578911924,83,721,1,805,10.372670807453416,community,1637,12.154781753927743
 openchat-v2-w-13b,9.615344158447204,0.8908241710735803,67,736,2,805,8.4472049689441,community,1566,12.03042777097436
+tulu-2-dpo-7b-ExPO,11.529221038762385,1.049781489308991,91,714,0,805,11.304347826086957,community,1742,11.675059099417426
 tulu-2-dpo-13b,10.119788388347828,0.929813366016608,75,728,2,805,9.440993788819876,community,1614,11.554479428088396
 claude2-alpaca-13b,7.437351324770187,0.82494288683272,59,746,0,805,7.329192546583851,community,1127,11.498898213160734
 minotaur-13b,5.738963669079602,0.7271241247374951,42,758,4,804,5.472636815920398,community,881,11.46525131683203

src/alpaca_eval/metrics/weights/weighted_alpaca_eval_gpt4_turbo/length_controlled_v1/baseline_gpt4_1106_preview.csv

@@ -146,3 +146,13 @@ Qwen1.5-1.8B-Chat,-1.6003884852505712,0.9646855557741588,-4.6744303356917447
 Qwen1.5-110B-Chat,-1.4481674391207744,0.9102999775192784,-0.2004892206655888
 Storm-7B,-0.3778112670657819,0.5727965213879709,0.0000000000000000
 SPPO-Mistral7B-PairRM,-1.0066475422582106,0.9046614612887018,-0.9905877944582094
+SPPO-Mistral7B-PairRM-ExPO,-0.9297137384620632,0.7671267711136246,-0.8709792439039323
+internlm2-chat-7b-ExPO,-1.1989304003616963,0.6968622384940820,-1.4260629123293445
+internlm2-chat-20b-ExPO,-1.5168999523223183,0.5813753405809725,-1.0854248669923878
+zephyr-7b-alpha-ExPO,-1.3654376133685404,1.1850496972307665,-2.7054511544809814
+zephyr-7b-beta-ExPO,-1.2721841274702139,0.7693453406014726,-2.2016614295073751
+Starling-LM-7B-alpha-ExPO,-1.1551552913433458,0.5427299165644314,-1.5681233228102427
+Starling-LM-7B-beta-ExPO,-0.9995849824567026,0.8173555243885808,-1.2278737751496258
+tulu-2-dpo-7b-ExPO,-1.2867594242188669,0.6986013668741516,-2.3831041176798933
+tulu-2-dpo-13b-ExPO,-1.6247537554410800,0.6431373083501301,-1.7734311638958129
+tulu-2-dpo-70b-ExPO,-1.2584665006823457,0.4518829275713181,-1.1294862478814247

src/alpaca_eval/models_configs/SPPO-Mistral7B-PairRM-ExPO/configs.yaml

@@ -0,0 +1,19 @@
+SPPO-Mistral7B-PairRM-ExPO:
+  prompt_template: "SPPO-Mistral7B-PairRM/prompt.txt"
+  fn_completions: "vllm_local_completions"
+  completions_kwargs:
+    model_name: "chujiezheng/Mistral7B-PairRM-SPPO-ExPO"
+    model_kwargs:
+      torch_dtype: 'bfloat16'
+      tokenizer_mode: 'auto'
+    max_new_tokens: 2048
+    do_sample: True
+    seed: 42
+    temperature: 0.7
+    top_k: 50
+    top_p: 0.9
+    presence_penalty: 0.1
+    frequency_penalty: 0.1
+    batch_size: 1000
+  pretty_name: "ExPO + SPPO-Mistral7B-PairRM"
+  link: "https://huggingface.co/chujiezheng/Mistral7B-PairRM-SPPO-ExPO"

src/alpaca_eval/models_configs/Starling-LM-7B-alpha-ExPO/configs.yaml

@@ -0,0 +1,19 @@
+Starling-LM-7B-alpha-ExPO:
+  prompt_template: "Starling-LM-7B-alpha/prompt.txt"
+  fn_completions: "vllm_local_completions"
+  completions_kwargs:
+    model_name: "chujiezheng/Starling-LM-7B-alpha-ExPO"
+    model_kwargs:
+      torch_dtype: 'bfloat16'
+      tokenizer_mode: 'auto'
+    max_new_tokens: 2048
+    do_sample: True
+    seed: 42
+    temperature: 0.7
+    top_k: 50
+    top_p: 0.9
+    presence_penalty: 0.1
+    frequency_penalty: 0.1
+    batch_size: 1000
+  pretty_name: "ExPO + Starling LM 7B alpha"
+  link: "https://huggingface.co/chujiezheng/Starling-LM-7B-alpha-ExPO"

src/alpaca_eval/models_configs/Starling-LM-7B-beta-ExPO/configs.yaml

@@ -0,0 +1,19 @@
+Starling-LM-7B-beta-ExPO:
+  prompt_template: "Starling-LM-7B-alpha/prompt.txt"
+  fn_completions: "vllm_local_completions"
+  completions_kwargs:
+    model_name: "chujiezheng/Starling-LM-7B-beta-ExPO"
+    model_kwargs:
+      torch_dtype: 'bfloat16'
+      tokenizer_mode: 'auto'
+    max_new_tokens: 2048
+    do_sample: True
+    seed: 42
+    temperature: 0.7
+    top_k: 50
+    top_p: 0.9
+    presence_penalty: 0.1
+    frequency_penalty: 0.1
+    batch_size: 1000
+  pretty_name: "ExPO + Starling LM 7B beta"
+  link: "https://huggingface.co/chujiezheng/Starling-LM-7B-beta-ExPO"

src/alpaca_eval/models_configs/internlm2-chat-20b-ExPO/configs.yaml

@@ -0,0 +1,20 @@
+internlm2-chat-20b-ExPO:
+  prompt_template: "internlm2-chat-20b-ppo/prompt.txt"
+  fn_completions: "vllm_local_completions"
+  completions_kwargs:
+    model_name: "chujiezheng/internlm2-chat-20b-ExPO"
+    model_kwargs:
+      torch_dtype: "bfloat16"
+      trust_remote_code: True
+      tokenizer_mode: 'auto'
+    max_new_tokens: 2048
+    do_sample: True
+    seed: 42
+    temperature: 0.7
+    top_k: 50
+    top_p: 0.9
+    presence_penalty: 0.1
+    frequency_penalty: 0.1
+    batch_size: 1000
+  pretty_name: "ExPO + InternLM2 Chat 20B"
+  link: "https://huggingface.co/chujiezheng/internlm2-chat-20b-ExPO"

src/alpaca_eval/models_configs/internlm2-chat-7b-ExPO/configs.yaml

@@ -0,0 +1,20 @@
+internlm2-chat-7b-ExPO:
+  prompt_template: "internlm2-chat-20b-ppo/prompt.txt"
+  fn_completions: "vllm_local_completions"
+  completions_kwargs:
+    model_name: "chujiezheng/internlm2-chat-7b-ExPO"
+    model_kwargs:
+      torch_dtype: "bfloat16"
+      trust_remote_code: True
+      tokenizer_mode: 'auto'
+    max_new_tokens: 2048
+    do_sample: True
+    seed: 42
+    temperature: 0.7
+    top_k: 50
+    top_p: 0.9
+    presence_penalty: 0.1
+    frequency_penalty: 0.1
+    batch_size: 1000
+  pretty_name: "ExPO + InternLM2 Chat 7B"
+  link: "https://huggingface.co/chujiezheng/internlm2-chat-7b-ExPO"

src/alpaca_eval/models_configs/tulu-2-dpo-13b-ExPO/configs.yaml

@@ -0,0 +1,19 @@
+tulu-2-dpo-13b-ExPO:
+  prompt_template: "tulu-2-dpo-70b/prompt.txt"
+  fn_completions: "vllm_local_completions"
+  completions_kwargs:
+    model_name: "chujiezheng/tulu-2-dpo-13b-ExPO"
+    model_kwargs:
+      torch_dtype: 'bfloat16'
+      tokenizer_mode: 'auto'
+    max_new_tokens: 2048
+    do_sample: True
+    seed: 42
+    temperature: 0.7
+    top_k: 50
+    top_p: 0.9
+    presence_penalty: 0.1
+    frequency_penalty: 0.1
+    batch_size: 1000
+  pretty_name: "ExPO + Tulu-2-DPO-13B"
+  link: "https://huggingface.co/chujiezheng/tulu-2-dpo-13b-ExPO"

src/alpaca_eval/models_configs/tulu-2-dpo-70b-ExPO/configs.yaml

@@ -0,0 +1,20 @@
+tulu-2-dpo-70b-ExPO:
+  prompt_template: "tulu-2-dpo-70b/prompt.txt"
+  fn_completions: "vllm_local_completions"
+  completions_kwargs:
+    model_name: "chujiezheng/tulu-2-dpo-70b-ExPO"
+    model_kwargs:
+      torch_dtype: 'bfloat16'
+      tokenizer_mode: 'auto'
+      tp: 2 # you need at least 2 A100 80GB GPUs to run this model
+    max_new_tokens: 2048
+    do_sample: True
+    seed: 42
+    temperature: 0.7
+    top_k: 50
+    top_p: 0.9
+    presence_penalty: 0.1
+    frequency_penalty: 0.1
+    batch_size: 1000
+  pretty_name: "ExPO + Tulu-2-DPO-70B"
+  link: "https://huggingface.co/chujiezheng/tulu-2-dpo-70b-ExPO"

src/alpaca_eval/models_configs/tulu-2-dpo-7b-ExPO/configs.yaml

@@ -0,0 +1,19 @@
+tulu-2-dpo-7b-ExPO:
+  prompt_template: "tulu-2-dpo-70b/prompt.txt"
+  fn_completions: "vllm_local_completions"
+  completions_kwargs:
+    model_name: "chujiezheng/tulu-2-dpo-7b-ExPO"
+    model_kwargs:
+      torch_dtype: 'bfloat16'
+      tokenizer_mode: 'auto'
+    max_new_tokens: 2048
+    do_sample: True
+    seed: 42
+    temperature: 0.7
+    top_k: 50
+    top_p: 0.9
+    presence_penalty: 0.1
+    frequency_penalty: 0.1
+    batch_size: 1000
+  pretty_name: "ExPO + Tulu-2-DPO-7B"
+  link: "https://huggingface.co/chujiezheng/tulu-2-dpo-7b-ExPO"

src/alpaca_eval/models_configs/zephyr-7b-alpha-ExPO/configs.yaml

@@ -0,0 +1,19 @@
+zephyr-7b-alpha-ExPO:
+  prompt_template: "zephyr-7b-alpha-ExPO/prompt.txt"
+  fn_completions: "vllm_local_completions"
+  completions_kwargs:
+    model_name: "chujiezheng/zephyr-7b-alpha-ExPO"
+    model_kwargs:
+      torch_dtype: 'bfloat16'
+      tokenizer_mode: 'auto'
+    max_new_tokens: 2048
+    do_sample: True
+    seed: 42
+    temperature: 0.7
+    top_k: 50
+    top_p: 0.9
+    presence_penalty: 0.1
+    frequency_penalty: 0.1
+    batch_size: 1000
+  pretty_name: "ExPO + Zephyr 7B Alpha"
+  link: "https://huggingface.co/HuggingFaceH4/zephyr-7b-alpha"

src/alpaca_eval/models_configs/zephyr-7b-alpha-ExPO/prompt.txt

@@ -0,0 +1,3 @@
+<|user|>
+{instruction}</s>
+<|assistant|>

src/alpaca_eval/models_configs/zephyr-7b-beta-ExPO/configs.yaml

@@ -0,0 +1,19 @@
+zephyr-7b-beta-ExPO:
+  prompt_template: "zephyr-7b-alpha-ExPO/prompt.txt"
+  fn_completions: "vllm_local_completions"
+  completions_kwargs:
+    model_name: "chujiezheng/zephyr-7b-beta-ExPO"
+    model_kwargs:
+      torch_dtype: 'bfloat16'
+      tokenizer_mode: 'auto'
+    max_new_tokens: 2048
+    do_sample: True
+    seed: 42
+    temperature: 0.7
+    top_k: 50
+    top_p: 0.9
+    presence_penalty: 0.1
+    frequency_penalty: 0.1
+    batch_size: 1000
+  pretty_name: "ExPO + Zephyr 7B Beta"
+  link: "https://huggingface.co/HuggingFaceH4/zephyr-7b-beta"