update

app.py

@@ -6,6 +6,7 @@ import pickle
 
 import gradio as gr
 import numpy as np
+import pandas as pd
 
 
 # notebook_url = "https://colab.research.google.com/drive/1RAWb22-PFNI-X1gPVzc927SGUdfr6nsR?usp=sharing"
@@ -16,17 +17,41 @@ basic_component_values = [None] * 6
 leader_component_values = [None] * 5
 
 
-def make_leaderboard_md(elo_results):
+def make_default_md(arena_df, elo_results):
+    total_votes = sum(arena_df["num_battles"]) // 2
+    total_models = len(arena_df)
+
     leaderboard_md = f"""
-# Leaderboard
-| [Vote](https://chat.lmsys.org/?arena) | [Blog](https://lmsys.org/blog/2023-05-03-arena/) | [GitHub](https://github.com/lm-sys/FastChat) | [Paper](https://arxiv.org/abs/2306.05685) | [Dataset](https://github.com/lm-sys/FastChat/blob/main/docs/dataset_release.md) | [Twitter](https://twitter.com/lmsysorg) | [Discord](https://discord.gg/HSWAKCrnFx) |
+# 🏆 LMSYS Chatbot Arena Leaderboard
+| [Vote](https://chat.lmsys.org) | [Blog](https://lmsys.org/blog/2023-05-03-arena/) | [GitHub](https://github.com/lm-sys/FastChat) | [Paper](https://arxiv.org/abs/2306.05685) | [Dataset](https://github.com/lm-sys/FastChat/blob/main/docs/dataset_release.md) | [Twitter](https://twitter.com/lmsysorg) | [Discord](https://discord.gg/HSWAKCrnFx) |
+
+LMSYS [Chatbot Arena](https://lmsys.org/blog/2023-05-03-arena/) is a crowdsourced open platform for LLM evals.
+We've collected over **200,000** human preference votes to rank LLMs with the Elo ranking system.
+"""
+    return leaderboard_md
+
 
-🏆 This leaderboard is based on the following three benchmarks.
-- [Chatbot Arena](https://chat.lmsys.org/?arena) - a crowdsourced, randomized battle platform. We use 130K+ user votes to compute Elo ratings.
-- [MT-Bench](https://arxiv.org/abs/2306.05685) - a set of challenging multi-turn questions. We use GPT-4 to grade the model responses.
-- [MMLU](https://arxiv.org/abs/2009.03300) (5-shot) - a test to measure a model's multitask accuracy on 57 tasks.
+def make_arena_leaderboard_md(arena_df):
+    total_votes = sum(arena_df["num_battles"]) // 2
+    total_models = len(arena_df)
 
-💻 Code: The Arena Elo ratings are computed by this [notebook]({notebook_url}). The MT-bench scores (single-answer grading on a scale of 10) are computed by [fastchat.llm_judge](https://github.com/lm-sys/FastChat/tree/main/fastchat/llm_judge). The MMLU scores are mostly computed by [InstructEval](https://github.com/declare-lab/instruct-eval). Higher values are better for all benchmarks. Empty cells mean not available. Last updated: Dec 20, 2023.
+    leaderboard_md = f"""
+Total #models: **{total_models}**. Total #votes: **{total_votes}**. Last updated: Jan 9, 2024.
+
+Contribute your vote 🗳️ at [chat.lmsys.org](https://chat.lmsys.org)! Find more analysis in the [notebook]({notebook_url}).
+"""
+    return leaderboard_md
+
+
+def make_full_leaderboard_md(elo_results):
+    leaderboard_md = f"""
+Two more benchmarks are displayed: **MT-Bench** and **MMLU**.
+- [MT-Bench](https://arxiv.org/abs/2306.05685): a set of challenging multi-turn questions. We use GPT-4 to grade the model responses.
+- [MMLU](https://arxiv.org/abs/2009.03300) (5-shot): a test to measure a model's multitask accuracy on 57 tasks.
+
+💻 Code: The MT-bench scores (single-answer grading on a scale of 10) are computed by [fastchat.llm_judge](https://github.com/lm-sys/FastChat/tree/main/fastchat/llm_judge).
+The MMLU scores are mostly computed by [InstructEval](https://github.com/declare-lab/instruct-eval).
+Higher values are better for all benchmarks. Empty cells mean not available.
 """
     return leaderboard_md
 
@@ -152,94 +177,190 @@ def build_basic_stats_tab():
             md4 = gr.Markdown(empty)
     return [md0, plot_1, md1, md2, md3, md4]
 
+def get_full_table(arena_df, model_table_df):
+    values = []
+    for i in range(len(model_table_df)):
+        row = []
+        model_key = model_table_df.iloc[i]["key"]
+        model_name = model_table_df.iloc[i]["Model"]
+        # model display name
+        row.append(model_name)
+        if model_key in arena_df.index:
+            idx = arena_df.index.get_loc(model_key)
+            row.append(round(arena_df.iloc[idx]["rating"], 1))
+        else:
+            row.append(np.nan)
+        row.append(model_table_df.iloc[i]["MT-bench (score)"])
+        row.append(model_table_df.iloc[i]["MMLU"])
+        # Organization
+        row.append(model_table_df.iloc[i]["Organization"])
+        # license
+        row.append(model_table_df.iloc[i]["License"])
+
+        values.append(row)
+    values.sort(key=lambda x: -x[1] if not np.isnan(x[1]) else 1e9)
+    return values
+
+
+def get_arena_table(arena_df, model_table_df):
+    # sort by rating
+    arena_df = arena_df.sort_values(by=["rating"], ascending=False)
+    values = []
+    for i in range(len(arena_df)):
+        row = []
+        model_key = arena_df.index[i]
+        model_name = model_table_df[model_table_df["key"] == model_key]["Model"].values[
+            0
+        ]
 
+        # rank
+        row.append(i + 1)
+        # model display name
+        row.append(model_name)
+        # elo rating
+        row.append(round(arena_df.iloc[i]["rating"], 1))
+        upper_diff = round(
+            arena_df.iloc[i]["rating_q975"] - arena_df.iloc[i]["rating"], 1
+        )
+        lower_diff = round(
+            arena_df.iloc[i]["rating"] - arena_df.iloc[i]["rating_q025"], 1
+        )
+        row.append(f"+{upper_diff}/-{lower_diff}")
+        # num battles
+        row.append(round(arena_df.iloc[i]["num_battles"]))
+        # Organization
+        row.append(
+            model_table_df[model_table_df["key"] == model_key]["Organization"].values[0]
+        )
+        # license
+        row.append(
+            model_table_df[model_table_df["key"] == model_key]["License"].values[0]
+        )
+
+        values.append(row)
+    return values
 
-def build_leaderboard_tab(elo_results_file, leaderboard_table_file):
+def build_leaderboard_tab(elo_results_file, leaderboard_table_file, show_plot=False):
     if elo_results_file is None:  # Do live update
-        md = "Loading ..."
+        default_md = "Loading ..."
         p1 = p2 = p3 = p4 = None
     else:
         with open(elo_results_file, "rb") as fin:
             elo_results = pickle.load(fin)
 
-        md = make_leaderboard_md(elo_results)
         p1 = elo_results["win_fraction_heatmap"]
         p2 = elo_results["battle_count_heatmap"]
         p3 = elo_results["bootstrap_elo_rating"]
         p4 = elo_results["average_win_rate_bar"]
+        arena_df = elo_results["leaderboard_table_df"]
+        default_md = make_default_md(arena_df, elo_results)
 
-    md_1 = gr.Markdown(md, elem_id="leaderboard_markdown")
-
+    md_1 = gr.Markdown(default_md, elem_id="leaderboard_markdown")
     if leaderboard_table_file:
         data = load_leaderboard_table_csv(leaderboard_table_file)
-        headers = [
-            "Model",
-            "Arena Elo rating",
-            "MT-bench (score)",
-            "MMLU",
-            "License",
-        ]
-        values = []
-        for item in data:
-            row = []
-            for key in headers:
-                value = item[key]
-                row.append(value)
-            values.append(row)
-        values.sort(key=lambda x: -x[1] if not np.isnan(x[1]) else 1e9)
-
-        headers[1] = "⭐ " + headers[1]
-        headers[2] = "📈 " + headers[2]
-
-        gr.Dataframe(
-            headers=headers,
-            datatype=["markdown", "number", "number", "number", "str"],
-            value=values,
-            elem_id="leaderboard_dataframe",
-        )
-        gr.Markdown(
-            "If you want to see more models, please help us [add them](https://github.com/lm-sys/FastChat/blob/main/docs/arena.md#how-to-add-a-new-model).",
-            elem_id="leaderboard_markdown"
-        )
+        model_table_df = pd.DataFrame(data)
+
+        with gr.Tabs() as tabs:
+            # arena table
+            arena_table_vals = get_arena_table(arena_df, model_table_df)
+            with gr.Tab("Arena Elo", id=0):
+                md = make_arena_leaderboard_md(arena_df)
+                gr.Markdown(md, elem_id="leaderboard_markdown")
+                gr.Dataframe(
+                    headers=[
+                        "Rank",
+                        "🤖 Model",
+                        "⭐ Arena Elo",
+                        "📊 95% CI",
+                        "🗳️ Votes",
+                        "Organization",
+                        "License",
+                    ],
+                    datatype=[
+                        "str",
+                        "markdown",
+                        "number",
+                        "str",
+                        "number",
+                        "str",
+                        "str",
+                    ],
+                    value=arena_table_vals,
+                    elem_id="arena_leaderboard_dataframe",
+                    height=700,
+                    column_widths=[50, 200, 100, 100, 100, 150, 150],
+                    wrap=True,
+                )
+            with gr.Tab("Full Leaderboard", id=1):
+                md = make_full_leaderboard_md(elo_results)
+                gr.Markdown(md, elem_id="leaderboard_markdown")
+                full_table_vals = get_full_table(arena_df, model_table_df)
+                gr.Dataframe(
+                    headers=[
+                        "🤖 Model",
+                        "⭐ Arena Elo",
+                        "📈 MT-bench",
+                        "📚 MMLU",
+                        "Organization",
+                        "License",
+                    ],
+                    datatype=["markdown", "number", "number", "number", "str", "str"],
+                    value=full_table_vals,
+                    elem_id="full_leaderboard_dataframe",
+                    column_widths=[200, 100, 100, 100, 150, 150],
+                    height=700,
+                    wrap=True,
+                )
+        if not show_plot:
+            gr.Markdown(
+                """ ## Visit our [HF space](https://huggingface.co/spaces/lmsys/chatbot-arena-leaderboard) for more analysis!
+                If you want to see more models, please help us [add them](https://github.com/lm-sys/FastChat/blob/main/docs/arena.md#how-to-add-a-new-model).
+                """,
+                elem_id="leaderboard_markdown",
+            )
     else:
         pass
 
-    gr.Markdown(
-        f"""## More Statistics for Chatbot Arena\n
-We added some additional figures to show more statistics. The code for generating them is also included in this [notebook]({notebook_url}).
-Please note that you may see different orders from different ranking methods. This is expected for models that perform similarly, as demonstrated by the confidence interval in the bootstrap figure. Going forward, we prefer the classical Elo calculation because of its scalability and interpretability. You can find more discussions in this blog [post](https://lmsys.org/blog/2023-05-03-arena/).
-""",
-        elem_id="leaderboard_markdown"
-    )
-
-    leader_component_values[:] = [md, p1, p2, p3, p4]
+    leader_component_values[:] = [default_md, p1, p2, p3, p4]
 
-    with gr.Row():
-        with gr.Column():
-            gr.Markdown(
-                "#### Figure 1: Fraction of Model A Wins for All Non-tied A vs. B Battles"
-            )
-            plot_1 = gr.Plot(p1, show_label=False)
-        with gr.Column():
-            gr.Markdown(
-                "#### Figure 2: Battle Count for Each Combination of Models (without Ties)"
-            )
-            plot_2 = gr.Plot(p2, show_label=False)
-    with gr.Row():
-        with gr.Column():
-            gr.Markdown(
-                "#### Figure 3: Bootstrap of MLE Elo Estimates (1000 Rounds of Random Sampling)"
-            )
-            plot_3 = gr.Plot(p3, show_label=False)
-        with gr.Column():
-            gr.Markdown(
-                "#### Figure 4: Average Win Rate Against All Other Models (Assuming Uniform Sampling and No Ties)"
-            )
-            plot_4 = gr.Plot(p4, show_label=False)
+    if show_plot:
+        gr.Markdown(
+            f"""## More Statistics for Chatbot Arena\n
+Below are figures for more statistics. The code for generating them is also included in this [notebook]({notebook_url}).
+You can find more discussions in this blog [post](https://lmsys.org/blog/2023-12-07-leaderboard/).
+    """,
+            elem_id="leaderboard_markdown"
+        )
+        with gr.Row():
+            with gr.Column():
+                gr.Markdown(
+                    "#### Figure 1: Fraction of Model A Wins for All Non-tied A vs. B Battles"
+                )
+                plot_1 = gr.Plot(p1, show_label=False)
+            with gr.Column():
+                gr.Markdown(
+                    "#### Figure 2: Battle Count for Each Combination of Models (without Ties)"
+                )
+                plot_2 = gr.Plot(p2, show_label=False)
+        with gr.Row():
+            with gr.Column():
+                gr.Markdown(
+                    "#### Figure 3: Bootstrap of Elo Estimates (1000 Rounds of Random Sampling)"
+                )
+                plot_3 = gr.Plot(p3, show_label=False)
+            with gr.Column():
+                gr.Markdown(
+                    "#### Figure 4: Average Win Rate Against All Other Models (Assuming Uniform Sampling and No Ties)"
+                )
+                plot_4 = gr.Plot(p4, show_label=False)
+
+    from fastchat.serve.gradio_web_server import acknowledgment_md
 
     gr.Markdown(acknowledgment_md)
 
-    return [md_1, plot_1, plot_2, plot_3, plot_4]
+    if show_plot:
+        return [md_1, plot_1, plot_2, plot_3, plot_4]
+    return [md_1]
 
 block_css = """
 #notice_markdown {
@@ -300,7 +421,7 @@ def build_demo(elo_results_file, leaderboard_table_file):
         css=block_css,
     ) as demo:
         leader_components = build_leaderboard_tab(
-            elo_results_file, leaderboard_table_file
+            elo_results_file, leaderboard_table_file, show_plot=True
         )
     return demo
 

elo_results_20240109.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3a334a1a5000f62dd9491d6fb2c7b136cce3fd37647ffec0e9c0c084919783ea
+size 264666

leaderboard_table_20240109.csv

@@ -0,0 +1,69 @@
+key,Model,MT-bench (score),MMLU,License,Organization,Link
+wizardlm-30b,WizardLM-30B,7.01,0.587,Non-commercial,Microsoft,https://huggingface.co/WizardLM/WizardLM-30B-V1.0
+vicuna-13b-16k,Vicuna-13B-16k,6.92,0.545,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,Non-commercial,Microsoft,https://huggingface.co/WizardLM/WizardLM-13B-V1.1
+tulu-30b,Tulu-30B,6.43,0.581,Non-commercial,AllenAI/UW,https://huggingface.co/allenai/tulu-30b
+guanaco-65b,Guanaco-65B,6.41,0.621,Non-commercial,UW,https://huggingface.co/timdettmers/guanaco-65b-merged
+openassistant-llama-30b,OpenAssistant-LLaMA-30B,6.41,0.560,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,Non-commercial,Microsoft,https://huggingface.co/WizardLM/WizardLM-13B-V1.0
+vicuna-7b-16k,Vicuna-7B-16k,6.22,0.485,Llama 2 Community,LMSYS,https://huggingface.co/lmsys/vicuna-7b-v1.5-16k
+baize-v2-13b,Baize-v2-13B,5.75,0.489,Non-commercial,UCSD,https://huggingface.co/project-baize/baize-v2-13b
+xgen-7b-8k-inst,XGen-7B-8K-Inst,5.55,0.421,Non-commercial,Salesforce,https://huggingface.co/Salesforce/xgen-7b-8k-inst
+nous-hermes-13b,Nous-Hermes-13B,5.51,0.493,Non-commercial,NousResearch,https://huggingface.co/NousResearch/Nous-Hermes-13b
+mpt-30b-instruct,MPT-30B-Instruct,5.22,0.478,CC-BY-SA 3.0,MosaicML,https://huggingface.co/mosaicml/mpt-30b-instruct
+falcon-40b-instruct,Falcon-40B-Instruct,5.17,0.547,Apache 2.0,TII,https://huggingface.co/tiiuae/falcon-40b-instruct
+h2o-oasst-openllama-13b,H2O-Oasst-OpenLLaMA-13B,4.63,0.428,Apache 2.0,h2oai,https://huggingface.co/h2oai/h2ogpt-gm-oasst1-en-2048-open-llama-13b
+gpt-4-turbo,GPT-4-Turbo,9.32,-,Proprietary,OpenAI,https://openai.com/blog/new-models-and-developer-products-announced-at-devday
+gpt-4-0314,GPT-4-0314,8.96,0.864,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,-,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,-,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,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,Proprietary,OpenAI,https://platform.openai.com/docs/models/gpt-3-5
+tulu-2-dpo-70b,Tulu-2-DPO-70B,7.89,-,AI2 ImpACT Low-risk,AllenAI/UW,https://huggingface.co/allenai/tulu-2-dpo-70b
+yi-34b-chat,Yi-34B-Chat,-,0.735,Yi License,01 AI,https://huggingface.co/01-ai/Yi-34B-Chat
+gemini-pro,Gemini Pro,-,0.718,Proprietary,Google,https://cloud.google.com/vertex-ai/docs/generative-ai/start/quickstarts/quickstart-multimodal
+gemini-pro-dev-api,Gemini Pro (Dev),-,0.718,Proprietary,Google,https://ai.google.dev/docs/gemini_api_overview
+wizardlm-70b,WizardLM-70B-v1.0,7.71,0.637,Llama 2 Community,Microsoft,https://huggingface.co/WizardLM/WizardLM-70B-V1.0
+vicuna-33b,Vicuna-33B,7.12,0.592,Non-commercial,LMSYS,https://huggingface.co/lmsys/vicuna-33b-v1.3
+starling-lm-7b-alpha,Starling-LM-7B-alpha,8.09,0.639,CC-BY-NC-4.0,UC Berkeley,https://huggingface.co/berkeley-nest/Starling-LM-7B-alpha
+pplx-70b-online,pplx-70b-online,-,-,Proprietary,Perplexity AI,https://blog.perplexity.ai/blog/introducing-pplx-online-llms
+openchat-3.5,OpenChat-3.5,7.81,0.643,Apache-2.0,OpenChat,https://huggingface.co/openchat/openchat_3.5
+openhermes-2.5-mistral-7b,OpenHermes-2.5-Mistral-7b,-,-,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,-,Proprietary,OpenAI,https://platform.openai.com/docs/models/gpt-3-5
+llama-2-70b-chat,Llama-2-70b-chat,6.86,0.630,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,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,-,-,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,Llama 2 Community,Microsoft,https://huggingface.co/WizardLM/WizardLM-13B-V1.2
+zephyr-7b-beta,Zephyr-7b-beta,7.34,0.614,MIT,HuggingFace,https://huggingface.co/HuggingFaceH4/zephyr-7b-beta
+mpt-30b-chat,MPT-30B-chat,6.39,0.504,CC-BY-NC-SA-4.0,MosaicML,https://huggingface.co/mosaicml/mpt-30b-chat
+vicuna-13b,Vicuna-13B,6.57,0.558,Llama 2 Community,LMSYS,https://huggingface.co/lmsys/vicuna-13b-v1.5
+qwen-14b-chat,Qwen-14B-Chat,6.96,0.665,Qianwen LICENSE,Alibaba,https://huggingface.co/Qwen/Qwen-14B-Chat
+zephyr-7b-alpha,Zephyr-7b-alpha,6.88,-,MIT,HuggingFace,https://huggingface.co/HuggingFaceH4/zephyr-7b-alpha
+codellama-34b-instruct,CodeLlama-34B-instruct,-,0.537,Llama 2 Community,Meta,https://huggingface.co/codellama/CodeLlama-34b-Instruct-hf
+falcon-180b-chat,falcon-180b-chat,-,0.680,Falcon-180B TII License,TII,https://huggingface.co/tiiuae/falcon-180B-chat
+guanaco-33b,Guanaco-33B,6.53,0.576,Non-commercial,UW,https://huggingface.co/timdettmers/guanaco-33b-merged
+llama-2-13b-chat,Llama-2-13b-chat,6.65,0.536,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,Apache 2.0,Mistral,https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1
+pplx-7b-online,pplx-7b-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,Llama 2 Community,Meta,https://huggingface.co/meta-llama/Llama-2-7b-chat-hf
+vicuna-7b,Vicuna-7B,6.17,0.498,Llama 2 Community,LMSYS,https://huggingface.co/lmsys/vicuna-7b-v1.5
+palm-2,PaLM-Chat-Bison-001,6.40,-,Proprietary,Google,https://cloud.google.com/vertex-ai/docs/generative-ai/learn/models#foundation_models
+koala-13b,Koala-13B,5.35,0.447,Non-commercial,UC Berkeley,https://bair.berkeley.edu/blog/2023/04/03/koala/
+chatglm3-6b,ChatGLM3-6B,-,-,Apache-2.0,Tsinghua,https://huggingface.co/THUDM/chatglm3-6b
+gpt4all-13b-snoozy,GPT4All-13B-Snoozy,5.41,0.430,Non-commercial,Nomic AI,https://huggingface.co/nomic-ai/gpt4all-13b-snoozy
+mpt-7b-chat,MPT-7B-Chat,5.42,0.320,CC-BY-NC-SA-4.0,MosaicML,https://huggingface.co/mosaicml/mpt-7b-chat
+chatglm2-6b,ChatGLM2-6B,4.96,0.455,Apache-2.0,Tsinghua,https://huggingface.co/THUDM/chatglm2-6b
+RWKV-4-Raven-14B,RWKV-4-Raven-14B,3.98,0.256,Apache 2.0,RWKV,https://huggingface.co/BlinkDL/rwkv-4-raven
+alpaca-13b,Alpaca-13B,4.53,0.481,Non-commercial,Stanford,https://crfm.stanford.edu/2023/03/13/alpaca.html
+oasst-pythia-12b,OpenAssistant-Pythia-12B,4.32,0.270,Apache 2.0,OpenAssistant,https://huggingface.co/OpenAssistant/oasst-sft-4-pythia-12b-epoch-3.5
+chatglm-6b,ChatGLM-6B,4.50,0.361,Non-commercial,Tsinghua,https://huggingface.co/THUDM/chatglm-6b
+fastchat-t5-3b,FastChat-T5-3B,3.04,0.477,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,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,MIT,Databricks,https://huggingface.co/databricks/dolly-v2-12b
+llama-13b,LLaMA-13B,2.61,0.470,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,Llama2-70B-SteerLM-Chat,7.54,-,Llama 2 Community,Nvidia,https://huggingface.co/nvidia/Llama2-70B-SteerLM-Chat