update

app.py

@@ -211,17 +211,25 @@ def create_ranking_str(ranking, ranking_difference):
     if ranking_difference > 0:
         return f"{int(ranking)} (\u2191{int(ranking_difference)})"
     elif ranking_difference < 0:
-        return f"{int(ranking)} ({int(-ranking_difference)}\u2193)"
+        return f"{int(ranking)} (\u2193{int(-ranking_difference)})"
     else:
         return f"{int(ranking)}"
     
 def get_arena_table(arena_df, model_table_df, arena_subset_df=None):
     arena_df = arena_df.sort_values(by=["rating"], ascending=False)
-    arena_df = arena_df.sort_values(by=["final_ranking"], ascending=True)
+    # arena_df = arena_df.sort_values(by=["final_ranking"], ascending=True)
+    arena_df = arena_df[arena_df["num_battles"] > 2000]
+
+    arena_df["final_ranking"] = range(1, len(arena_df) + 1)
     # sort by rating
-    if arena_subset_df is not None: 
+    if arena_subset_df is not None:
+        # filter out models not in the arena_df
+        arena_subset_df = arena_subset_df[arena_subset_df.index.isin(arena_df.index)]
         arena_subset_df = arena_subset_df.sort_values(by=["rating"], ascending=False)
-        arena_subset_df = arena_subset_df.sort_values(by=["final_ranking"], ascending=True)
+        # arena_subset_df = arena_subset_df.sort_values(by=["final_ranking"], ascending=True)
+
+        # assign ranking by the order
+        arena_subset_df["final_ranking"] = range(1, len(arena_subset_df) + 1)
         # join arena_df and arena_subset_df on index
         arena_df = arena_subset_df.join(arena_df["final_ranking"], rsuffix="_global", how="inner")
         arena_df['ranking_difference'] =  arena_df['final_ranking_global'] - arena_df['final_ranking']
@@ -294,6 +302,7 @@ def build_leaderboard_tab(elo_results_file, leaderboard_table_file, show_plot=Fa
                 elo_chinese_results = elo_results["chinese"]
                 elo_long_results = elo_results["long"]
                 elo_english_results = elo_results["english"]
+                elo_coding_results = elo_results["coding"]
                 elo_results = elo_results["full"]
 
         p1 = elo_results["win_fraction_heatmap"]
@@ -304,6 +313,7 @@ def build_leaderboard_tab(elo_results_file, leaderboard_table_file, show_plot=Fa
         arena_chinese_df = elo_chinese_results["leaderboard_table_df"]
         arena_long_df = elo_long_results["leaderboard_table_df"]
         arena_english_df = elo_english_results["leaderboard_table_df"]
+        arena_coding_df = elo_coding_results["leaderboard_table_df"]
         default_md = make_default_md(arena_df, elo_results)
 
     md_1 = gr.Markdown(default_md, elem_id="leaderboard_markdown")
@@ -322,15 +332,17 @@ def build_leaderboard_tab(elo_results_file, leaderboard_table_file, show_plot=Fa
                     overall_rating = gr.Button("Overall")
                     # update_overall_rating_df = lambda _: get_arena_table(arena_df, model_table_df)
                     update_overall_rating_df = lambda x: update_leaderboard_and_plots(x, arena_df, model_table_df, None, elo_results)
+                    coding_rating = gr.Button("Coding")
+                    update_coding_rating_df = lambda x: update_leaderboard_and_plots(x, arena_df, model_table_df, arena_coding_df, elo_coding_results)
+                    long_context_rating = gr.Button("Long Conversation")
+                    update_long_context_rating_df = lambda x: update_leaderboard_and_plots(x, arena_df, model_table_df, arena_long_df, elo_long_results)
+                    # update_long_context_rating_df = lambda _: get_arena_table(arena_df, model_table_df, arena_long_df)
                     english_rating = gr.Button("English")
                     update_english_rating_df = lambda x: update_leaderboard_and_plots(x, arena_df, model_table_df, arena_english_df, elo_english_results)
                     # update_english_rating_df = lambda _: get_arena_table(arena_df, model_table_df, arena_english_df)
                     chinese_rating = gr.Button("Chinese")
                     update_chinese_rating_df = lambda x: update_leaderboard_and_plots(x, arena_df, model_table_df, arena_chinese_df, elo_chinese_results)
                     # update_chinese_rating_df = lambda _: get_arena_table(arena_df, model_table_df, arena_chinese_df)
-                    long_context_rating = gr.Button("Long Context")
-                    update_long_context_rating_df = lambda x: update_leaderboard_and_plots(x, arena_df, model_table_df, arena_long_df, elo_long_results)
-                    # update_long_context_rating_df = lambda _: get_arena_table(arena_df, model_table_df, arena_long_df)
                 elo_display_df = gr.Dataframe(
                     headers=[
                         "Rank",
@@ -428,9 +440,10 @@ See Figure 3 below for visualization of the confidence intervals. Code to recrea
                 plot_4 = gr.Plot(p4, show_label=False)
                 
     overall_rating.click(fn=update_overall_rating_df, inputs=overall_rating, outputs=[elo_display_df, plot_1, plot_2, plot_3, plot_4, more_stats_md, leaderboard_markdown])
+    coding_rating.click(fn=update_coding_rating_df, inputs=coding_rating, outputs=[elo_display_df, plot_1, plot_2, plot_3, plot_4, more_stats_md, leaderboard_markdown])
+    long_context_rating.click(fn=update_long_context_rating_df, inputs=long_context_rating, outputs=[elo_display_df, plot_1, plot_2, plot_3, plot_4, more_stats_md, leaderboard_markdown])
     english_rating.click(fn=update_english_rating_df, inputs=english_rating, outputs=[elo_display_df, plot_1, plot_2, plot_3, plot_4, more_stats_md, leaderboard_markdown])
     chinese_rating.click(fn=update_chinese_rating_df, inputs=chinese_rating ,outputs=[elo_display_df, plot_1, plot_2, plot_3, plot_4, more_stats_md, leaderboard_markdown])
-    long_context_rating.click(fn=update_long_context_rating_df, inputs=long_context_rating, outputs=[elo_display_df, plot_1, plot_2, plot_3, plot_4, more_stats_md, leaderboard_markdown])
 
     with gr.Accordion(
         "📝 Citation",
@@ -551,6 +564,8 @@ def build_demo(elo_results_file, leaderboard_table_file):
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()
     parser.add_argument("--share", action="store_true")
+    parser.add_argument("--host", default="0.0.0.0")
+    parser.add_argument("--port", type=int, default=7860)
     args = parser.parse_args()
 
     elo_result_files = glob.glob("elo_results_*.pkl")
@@ -562,4 +577,4 @@ if __name__ == "__main__":
     leaderboard_table_file = leaderboard_table_files[-1]
 
     demo = build_demo(elo_result_file, leaderboard_table_file)
-    demo.launch(share=args.share)
+    demo.launch(share=args.share, server_name=args.host, server_port=args.port)

leaderboard_table_20240404.csv

@@ -0,0 +1,94 @@
+key,Model,MT-bench (score),MMLU,Knowledge cutoff date,License,Organization,Link
+wizardlm-30b,WizardLM-30B,7.01,0.587,2023/6,Non-commercial,Microsoft,https://huggingface.co/WizardLM/WizardLM-30B-V1.0
+vicuna-13b-16k,Vicuna-13B-16k,6.92,0.545,2023/7,Llama 2 Community,LMSYS,https://huggingface.co/lmsys/vicuna-13b-v1.5-16k
+wizardlm-13b-v1.1,WizardLM-13B-v1.1,6.76,0.500,2023/7,Non-commercial,Microsoft,https://huggingface.co/WizardLM/WizardLM-13B-V1.1
+tulu-30b,Tulu-30B,6.43,0.581,2023/6,Non-commercial,AllenAI/UW,https://huggingface.co/allenai/tulu-30b
+guanaco-65b,Guanaco-65B,6.41,0.621,2023/5,Non-commercial,UW,https://huggingface.co/timdettmers/guanaco-65b-merged
+openassistant-llama-30b,OpenAssistant-LLaMA-30B,6.41,0.560,2023/4,Non-commercial,OpenAssistant,https://huggingface.co/OpenAssistant/oasst-sft-6-llama-30b-xor
+wizardlm-13b-v1.0,WizardLM-13B-v1.0,6.35,0.523,2023/5,Non-commercial,Microsoft,https://huggingface.co/WizardLM/WizardLM-13B-V1.0
+vicuna-7b-16k,Vicuna-7B-16k,6.22,0.485,2023/7,Llama 2 Community,LMSYS,https://huggingface.co/lmsys/vicuna-7b-v1.5-16k
+baize-v2-13b,Baize-v2-13B,5.75,0.489,2023/4,Non-commercial,UCSD,https://huggingface.co/project-baize/baize-v2-13b
+xgen-7b-8k-inst,XGen-7B-8K-Inst,5.55,0.421,2023/7,Non-commercial,Salesforce,https://huggingface.co/Salesforce/xgen-7b-8k-inst
+nous-hermes-13b,Nous-Hermes-13B,5.51,0.493,2023/6,Non-commercial,NousResearch,https://huggingface.co/NousResearch/Nous-Hermes-13b
+mpt-30b-instruct,MPT-30B-Instruct,5.22,0.478,2023/6,CC-BY-SA 3.0,MosaicML,https://huggingface.co/mosaicml/mpt-30b-instruct
+falcon-40b-instruct,Falcon-40B-Instruct,5.17,0.547,2023/5,Apache 2.0,TII,https://huggingface.co/tiiuae/falcon-40b-instruct
+h2o-oasst-openllama-13b,H2O-Oasst-OpenLLaMA-13B,4.63,0.428,2023/6,Apache 2.0,h2oai,https://huggingface.co/h2oai/h2ogpt-gm-oasst1-en-2048-open-llama-13b
+gpt-4-1106-preview,GPT-4-1106-preview,9.32,-,2023/4,Proprietary,OpenAI,https://openai.com/blog/new-models-and-developer-products-announced-at-devday
+gpt-4-0314,GPT-4-0314,8.96,0.864,2021/9,Proprietary,OpenAI,https://openai.com/research/gpt-4
+claude-1,Claude-1,7.90,0.770,-,Proprietary,Anthropic,https://www.anthropic.com/index/introducing-claude
+gpt-4-0613,GPT-4-0613,9.18,-,2021/9,Proprietary,OpenAI,https://platform.openai.com/docs/models/gpt-4-and-gpt-4-turbo
+claude-2.0,Claude-2.0,8.06,0.785,-,Proprietary,Anthropic,https://www.anthropic.com/index/claude-2
+claude-2.1,Claude-2.1,8.18,-,-,Proprietary,Anthropic,https://www.anthropic.com/index/claude-2-1
+gpt-3.5-turbo-0613,GPT-3.5-Turbo-0613,8.39,-,2021/9,Proprietary,OpenAI,https://platform.openai.com/docs/models/gpt-3-5
+mixtral-8x7b-instruct-v0.1,Mixtral-8x7b-Instruct-v0.1,8.30,0.706,2023/12,Apache 2.0,Mistral,https://mistral.ai/news/mixtral-of-experts/
+claude-instant-1,Claude-Instant-1,7.85,0.734,-,Proprietary,Anthropic,https://www.anthropic.com/index/introducing-claude
+gpt-3.5-turbo-0314,GPT-3.5-Turbo-0314,7.94,0.700,2021/9,Proprietary,OpenAI,https://platform.openai.com/docs/models/gpt-3-5
+tulu-2-dpo-70b,Tulu-2-DPO-70B,7.89,-,2023/11,AI2 ImpACT Low-risk,AllenAI/UW,https://huggingface.co/allenai/tulu-2-dpo-70b
+yi-34b-chat,Yi-34B-Chat,-,0.735,2023/6,Yi License,01 AI,https://huggingface.co/01-ai/Yi-34B-Chat
+gemini-pro,Gemini Pro,-,0.718,2023/4,Proprietary,Google,https://blog.google/technology/ai/gemini-api-developers-cloud/
+gemini-pro-dev-api,Gemini Pro (Dev API),-,0.718,2023/4,Proprietary,Google,https://ai.google.dev/docs/gemini_api_overview
+bard-jan-24-gemini-pro,Bard (Gemini Pro),-,-,Online,Proprietary,Google,https://bard.google.com/
+wizardlm-70b,WizardLM-70B-v1.0,7.71,0.637,2023/8,Llama 2 Community,Microsoft,https://huggingface.co/WizardLM/WizardLM-70B-V1.0
+vicuna-33b,Vicuna-33B,7.12,0.592,2023/8,Non-commercial,LMSYS,https://huggingface.co/lmsys/vicuna-33b-v1.3
+starling-lm-7b-alpha,Starling-LM-7B-alpha,8.09,0.639,2023/11,CC-BY-NC-4.0,UC Berkeley,https://huggingface.co/berkeley-nest/Starling-LM-7B-alpha
+pplx-70b-online,pplx-70b-online,-,-,Online,Proprietary,Perplexity AI,https://blog.perplexity.ai/blog/introducing-pplx-online-llms
+openchat-3.5,OpenChat-3.5,7.81,0.643,2023/11,Apache-2.0,OpenChat,https://huggingface.co/openchat/openchat_3.5
+openhermes-2.5-mistral-7b,OpenHermes-2.5-Mistral-7b,-,-,2023/11,Apache-2.0,NousResearch,https://huggingface.co/teknium/OpenHermes-2.5-Mistral-7B
+gpt-3.5-turbo-1106,GPT-3.5-Turbo-1106,8.32,-,2021/9,Proprietary,OpenAI,https://platform.openai.com/docs/models/gpt-3-5
+llama-2-70b-chat,Llama-2-70b-chat,6.86,0.630,2023/7,Llama 2 Community,Meta,https://huggingface.co/meta-llama/Llama-2-70b-chat-hf
+solar-10.7b-instruct-v1.0,SOLAR-10.7B-Instruct-v1.0,7.58,0.662,2023/11,CC-BY-NC-4.0,Upstage AI,https://huggingface.co/upstage/SOLAR-10.7B-Instruct-v1.0
+dolphin-2.2.1-mistral-7b,Dolphin-2.2.1-Mistral-7B,-,-,2023/10,Apache-2.0,Cognitive Computations,https://huggingface.co/ehartford/dolphin-2.2.1-mistral-7b
+wizardlm-13b,WizardLM-13b-v1.2,7.20,0.527,2023/7,Llama 2 Community,Microsoft,https://huggingface.co/WizardLM/WizardLM-13B-V1.2
+zephyr-7b-beta,Zephyr-7b-beta,7.34,0.614,2023/10,MIT,HuggingFace,https://huggingface.co/HuggingFaceH4/zephyr-7b-beta
+mpt-30b-chat,MPT-30B-chat,6.39,0.504,2023/6,CC-BY-NC-SA-4.0,MosaicML,https://huggingface.co/mosaicml/mpt-30b-chat
+vicuna-13b,Vicuna-13B,6.57,0.558,2023/7,Llama 2 Community,LMSYS,https://huggingface.co/lmsys/vicuna-13b-v1.5
+qwen-14b-chat,Qwen-14B-Chat,6.96,0.665,2023/8,Qianwen LICENSE,Alibaba,https://huggingface.co/Qwen/Qwen-14B-Chat
+zephyr-7b-alpha,Zephyr-7b-alpha,6.88,-,2023/10,MIT,HuggingFace,https://huggingface.co/HuggingFaceH4/zephyr-7b-alpha
+codellama-34b-instruct,CodeLlama-34B-instruct,-,0.537,2023/7,Llama 2 Community,Meta,https://huggingface.co/codellama/CodeLlama-34b-Instruct-hf
+falcon-180b-chat,falcon-180b-chat,-,0.680,2023/9,Falcon-180B TII License,TII,https://huggingface.co/tiiuae/falcon-180B-chat
+guanaco-33b,Guanaco-33B,6.53,0.576,2023/5,Non-commercial,UW,https://huggingface.co/timdettmers/guanaco-33b-merged
+llama-2-13b-chat,Llama-2-13b-chat,6.65,0.536,2023/7,Llama 2 Community,Meta,https://huggingface.co/meta-llama/Llama-2-13b-chat-hf
+mistral-7b-instruct,Mistral-7B-Instruct-v0.1,6.84,0.554,2023/9,Apache 2.0,Mistral,https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1
+pplx-7b-online,pplx-7b-online,-,-,Online,Proprietary,Perplexity AI,https://blog.perplexity.ai/blog/introducing-pplx-online-llms
+llama-2-7b-chat,Llama-2-7b-chat,6.27,0.458,2023/7,Llama 2 Community,Meta,https://huggingface.co/meta-llama/Llama-2-7b-chat-hf
+vicuna-7b,Vicuna-7B,6.17,0.498,2023/7,Llama 2 Community,LMSYS,https://huggingface.co/lmsys/vicuna-7b-v1.5
+palm-2,PaLM-Chat-Bison-001,6.40,-,2021/6,Proprietary,Google,https://cloud.google.com/vertex-ai/docs/generative-ai/learn/models#foundation_models
+koala-13b,Koala-13B,5.35,0.447,2023/4,Non-commercial,UC Berkeley,https://bair.berkeley.edu/blog/2023/04/03/koala/
+chatglm3-6b,ChatGLM3-6B,-,-,2023/10,Apache-2.0,Tsinghua,https://huggingface.co/THUDM/chatglm3-6b
+gpt4all-13b-snoozy,GPT4All-13B-Snoozy,5.41,0.430,2023/3,Non-commercial,Nomic AI,https://huggingface.co/nomic-ai/gpt4all-13b-snoozy
+mpt-7b-chat,MPT-7B-Chat,5.42,0.320,2023/5,CC-BY-NC-SA-4.0,MosaicML,https://huggingface.co/mosaicml/mpt-7b-chat
+chatglm2-6b,ChatGLM2-6B,4.96,0.455,2023/6,Apache-2.0,Tsinghua,https://huggingface.co/THUDM/chatglm2-6b
+RWKV-4-Raven-14B,RWKV-4-Raven-14B,3.98,0.256,2023/4,Apache 2.0,RWKV,https://huggingface.co/BlinkDL/rwkv-4-raven
+alpaca-13b,Alpaca-13B,4.53,0.481,2023/3,Non-commercial,Stanford,https://crfm.stanford.edu/2023/03/13/alpaca.html
+oasst-pythia-12b,OpenAssistant-Pythia-12B,4.32,0.270,2023/4,Apache 2.0,OpenAssistant,https://huggingface.co/OpenAssistant/oasst-sft-4-pythia-12b-epoch-3.5
+chatglm-6b,ChatGLM-6B,4.50,0.361,2023/3,Non-commercial,Tsinghua,https://huggingface.co/THUDM/chatglm-6b
+fastchat-t5-3b,FastChat-T5-3B,3.04,0.477,2023/4,Apache 2.0,LMSYS,https://huggingface.co/lmsys/fastchat-t5-3b-v1.0
+stablelm-tuned-alpha-7b,StableLM-Tuned-Alpha-7B,2.75,0.244,2023/4,CC-BY-NC-SA-4.0,Stability AI,https://huggingface.co/stabilityai/stablelm-tuned-alpha-7b
+dolly-v2-12b,Dolly-V2-12B,3.28,0.257,2023/4,MIT,Databricks,https://huggingface.co/databricks/dolly-v2-12b
+llama-13b,LLaMA-13B,2.61,0.470,2023/2,Non-commercial,Meta,https://arxiv.org/abs/2302.13971
+mistral-medium,Mistral Medium,8.61,0.753,-,Proprietary,Mistral,https://mistral.ai/news/la-plateforme/
+llama2-70b-steerlm-chat,NV-Llama2-70B-SteerLM-Chat,7.54,0.685,2023/11,Llama 2 Community,Nvidia,https://huggingface.co/nvidia/Llama2-70B-SteerLM-Chat
+stripedhyena-nous-7b,StripedHyena-Nous-7B,-,-,2023/12,Apache 2.0,Together AI,https://huggingface.co/togethercomputer/StripedHyena-Nous-7B
+deepseek-llm-67b-chat,DeepSeek-LLM-67B-Chat,-,0.713,2023/11,DeepSeek License,DeepSeek AI,https://huggingface.co/deepseek-ai/deepseek-llm-67b-chat
+gpt-4-0125-preview,GPT-4-0125-preview,-,-,2023/12,Proprietary,OpenAI,https://openai.com/blog/new-models-and-developer-products-announced-at-devday
+qwen1.5-72b-chat,Qwen1.5-72B-Chat,8.61,0.775,2024/2,Qianwen LICENSE,Alibaba,https://qwenlm.github.io/blog/qwen1.5/
+qwen1.5-7b-chat,Qwen1.5-7B-Chat,7.6,0.610,2024/2,Qianwen LICENSE,Alibaba,https://qwenlm.github.io/blog/qwen1.5/
+qwen1.5-4b-chat,Qwen1.5-4B-Chat,-,0.561,2024/2,Qianwen LICENSE,Alibaba,https://qwenlm.github.io/blog/qwen1.5/
+openchat-3.5-0106,OpenChat-3.5-0106,7.8,0.658,2024/1,Apache-2.0,OpenChat,https://huggingface.co/openchat/openchat-3.5-0106
+nous-hermes-2-mixtral-8x7b-dpo,Nous-Hermes-2-Mixtral-8x7B-DPO,-,-,2024/1,Apache-2.0,NousResearch,https://huggingface.co/NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO
+gpt-3.5-turbo-0125,GPT-3.5-Turbo-0125,-,-,2021/9,Proprietary,OpenAI,https://platform.openai.com/docs/models/gpt-3-5-turbo
+mistral-next,Mistral-Next,-,-,-,Proprietary,Mistral,https://chat.mistral.ai/chat
+mistral-large-2402,Mistral-Large-2402,-,0.812,-,Proprietary,Mistral,https://mistral.ai/news/mistral-large/
+gemma-7b-it,Gemma-7B-it,-,0.643,2024/2,Gemma license,Google,https://huggingface.co/google/gemma-7b-it
+gemma-2b-it,Gemma-2B-it,-,0.423,2024/2,Gemma license,Google,https://huggingface.co/google/gemma-2b-it
+mistral-7b-instruct-v0.2,Mistral-7B-Instruct-v0.2,7.6,-,2023/12,Apache-2.0,Mistral,https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2
+claude-3-sonnet-20240229,Claude 3 Sonnet,-,0.790,2023/8,Proprietary,Anthropic,https://www.anthropic.com/news/claude-3-family
+claude-3-opus-20240229,Claude 3 Opus,-,0.868,2023/8,Proprietary,Anthropic,https://www.anthropic.com/news/claude-3-family
+codellama-70b-instruct,CodeLlama-70B-instruct,-,-,2024/1,Llama 2 Community,Meta,https://huggingface.co/codellama/CodeLlama-70b-hf
+olmo-7b-instruct,OLMo-7B-instruct,-,-,2024/2,Apache-2.0,Allen AI,https://huggingface.co/allenai/OLMo-7B-Instruct
+claude-3-haiku-20240307,Claude 3 Haiku,-,0.752,2023/8,Proprietary,Anthropic,https://www.anthropic.com/news/claude-3-family
+starling-lm-7b-beta,Starling-LM-7B-beta,8.12,-,2024/3,Apache-2.0,Nexusflow,https://huggingface.co/Nexusflow/Starling-LM-7B-beta
+dbrx-instruct,DBRX-instruct,-,-,2024/3,Apache-2.0,Databricks,-
+command-r,Command R,-,-,2024/3,Apache-2.0,Cohere,-
+qwen1.5-14b-chat,Qwen1.5-14B-Chat,-,-,2024/2,Qianwen LICENSE,Alibaba,https://qwenlm.github.io/blog/qwen1.5/
+qwen1.5-32b-chat,Qwen1.5-32B-Chat,-,-,2024/2,Qianwen LICENSE,Alibaba,https://qwenlm.github.io/blog/qwen1.5/