init

.gitignore

@@ -0,0 +1,8 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+.env
+.ipynb_checkpoints
+*.pyc
+*ipynb
+.vscode/
+.idea/

README.md

@@ -1,13 +1,12 @@
 ---
 title: UnlearnCanvas Benchmark
-emoji: 🐨
-colorFrom: green
-colorTo: yellow
+emoji: 🎨
+colorFrom: red
+colorTo: purple
 sdk: gradio
-sdk_version: 4.19.1
+sdk_version: 4.17.0
 app_file: app.py
 pinned: false
-license: apache-2.0
 ---
 
 Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app.py

@@ -0,0 +1,501 @@
+import gradio as gr
+import pandas as pd
+from src.display.about import (
+    CITATION_BUTTON_LABEL,
+    CITATION_BUTTON_TEXT,
+    EVALUATION_QUEUE_TEXT,
+    INTRODUCTION_TEXT,
+    LLM_BENCHMARKS_TEXT,
+    FAQ_TEXT,
+    TITLE,
+)
+from src.display.css_html_js import custom_css
+from src.display.utils import (
+    BENCHMARK_COLS,
+    COLS,
+    EVAL_COLS,
+    EVAL_TYPES,
+    NUMERIC_INTERVALS,
+    TYPES,
+    AutoEvalColumn,
+    ModelType,
+    fields,
+    WeightType,
+    Precision
+)
+from src.envs import API, EVAL_REQUESTS_PATH, EVAL_RESULTS_PATH, H4_TOKEN, IS_PUBLIC, QUEUE_REPO, REPO_ID, RESULTS_REPO
+from PIL import Image
+# from src.populate import get_evaluation_queue_df, get_leaderboard_df
+# from src.submission.submit import add_new_eval
+# from src.tools.collections import update_collections
+# from src.tools.plots import (
+#     create_metric_plot_obj,
+#     create_plot_df,
+#     create_scores_df,
+# )
+from dummydatagen import dummy_data_for_plot, create_metric_plot_obj_1, dummydf
+import copy
+
+
+def restart_space():
+    API.restart_space(repo_id=REPO_ID, token=H4_TOKEN)
+
+
+def add_average_col(df):
+
+    always_here_cols = [
+        "Model", "Agent", "Opponent Model", "Opponent Agent"
+    ]
+    desired_col = [i for i in list(df.columns) if i not in always_here_cols]
+    newdf = df[desired_col].mean(axis=1).round(3)
+    return newdf
+
+
+gtbench_raw_data = dummydf()
+gtbench_raw_data["Average"] = add_average_col(gtbench_raw_data)
+
+column_to_move = "Average"
+# Move the column to the desired index
+gtbench_raw_data.insert(
+    4, column_to_move, gtbench_raw_data.pop(column_to_move))
+
+models = list(set(gtbench_raw_data['Model']))
+
+opponent_models = list(set(gtbench_raw_data['Opponent Model']))
+
+
+agents = list(set(gtbench_raw_data['Agent']))
+
+
+opponent_agents = list(set(gtbench_raw_data['Opponent Agent']))
+
+# Searching and filtering
+
+
+def update_table(
+    hidden_df: pd.DataFrame,
+    columns: list,
+    model1: list,
+    model2: list,
+    agent1: list,
+    agent2: list
+):
+
+    filtered_df = select_columns(hidden_df, columns)
+
+    filtered_df = filter_model1(filtered_df, model1)
+    filtered_df = filter_model2(filtered_df, model2)
+    filtered_df = filter_agent1(filtered_df, agent1)
+    filtered_df = filter_agent2(filtered_df, agent2)
+
+    return filtered_df
+
+# triggered only once at startup => read query parameter if it exists
+
+
+def load_query(request: gr.Request):
+    query = request.query_params.get("query") or ""
+    return query, query  # return one for the "search_bar", one for a hidden component that triggers a reload only if value has changed
+
+
+def search_table(df: pd.DataFrame, query: str) -> pd.DataFrame:
+    return df[(df[AutoEvalColumn.dummy.name].str.contains(query, case=False))]
+
+
+def select_columns(df: pd.DataFrame, columns: list) -> pd.DataFrame:
+    always_here_cols = [
+        "Model", "Agent", "Opponent Model", "Opponent Agent"
+    ]
+    # We use COLS to maintain sorting
+    all_columns = games
+
+    if len(columns) == 0:
+        filtered_df = df[
+            always_here_cols +
+            [c for c in all_columns if c in df.columns]
+        ]
+        filtered_df["Average"] = add_average_col(filtered_df)
+        column_to_move = "Average"
+        current_index = filtered_df.columns.get_loc(column_to_move)
+
+        # Move the column to the desired index
+        filtered_df.insert(4, column_to_move, filtered_df.pop(column_to_move))
+        return filtered_df
+
+    filtered_df = df[
+        always_here_cols +
+        [c for c in all_columns if c in df.columns and c in columns]
+    ]
+    if "Average" in columns:
+        filtered_df["Average"] = add_average_col(filtered_df)
+        # Get the current index of the column
+        column_to_move = "Average"
+        current_index = filtered_df.columns.get_loc(column_to_move)
+
+        # Move the column to the desired index
+        filtered_df.insert(4, column_to_move, filtered_df.pop(column_to_move))
+    else:
+        if "Average" in filtered_df.columns:
+            # Remove the column
+            filtered_df = filtered_df.drop(columns=["Average"])
+
+    return filtered_df
+
+
+def filter_model1(
+    df: pd.DataFrame, model_query: list
+) -> pd.DataFrame:
+    # Show all models
+    if len(model_query) == 0:
+        return df
+    filtered_df = df
+
+    filtered_df = filtered_df[filtered_df["Model"].isin(
+        model_query)]
+    return filtered_df
+
+
+def filter_model2(
+    df: pd.DataFrame, model_query: list
+) -> pd.DataFrame:
+    # Show all models
+    if len(model_query) == 0:
+        return df
+    filtered_df = df
+
+    filtered_df = filtered_df[filtered_df["Opponent Model"].isin(
+        model_query)]
+    return filtered_df
+
+
+def filter_agent1(
+    df: pd.DataFrame, agent_query: list
+) -> pd.DataFrame:
+    # Show all models
+    if len(agent_query) == 0:
+        return df
+    filtered_df = df
+
+    filtered_df = filtered_df[filtered_df["Agent"].isin(
+        agent_query)]
+    return filtered_df
+
+
+def filter_agent2(
+    df: pd.DataFrame, agent_query: list
+) -> pd.DataFrame:
+    # Show all models
+    if len(agent_query) == 0:
+        return df
+    filtered_df = df
+
+    filtered_df = filtered_df[filtered_df["Opponent Agent"].isin(
+        agent_query)]
+    return filtered_df
+
+
+# leaderboard_df = filter_models(leaderboard_df, [t.to_str(" : ") for t in ModelType], list(NUMERIC_INTERVALS.keys()), [i.value.name for i in Precision], False, False)
+
+
+class LLM_Model:
+    def __init__(self, t_value, model_value, average_value, arc_value, hellaSwag_value, mmlu_value) -> None:
+        self.t = t_value
+        self.model = model_value
+        self.average = average_value
+        self.arc = arc_value
+        self.hellaSwag = hellaSwag_value
+        self.mmlu = mmlu_value
+
+
+games = ["Breakthrough", "Connect Four", "Blind Auction", "Kuhn Poker",
+         "Liar's Dice", "Negotiation", "Nim", "Pig", "Iterated Prisoner's Dilemma", "Tic-Tac-Toe"]
+
+# models = ["gpt-35-turbo-1106", "gpt-4", "Llama-2-70b-chat-hf", "CodeLlama-34b-Instruct-hf",
+#           "CodeLlama-70b-Instruct-hf", "Mistral-7B-Instruct-v01", "Mistral-7B-OpenOrca"]
+
+# agents = ["Prompt Agent", "CoT Agent", "SC-CoT Agent",
+#           "ToT Agent", "MCTS", "Random", "TitforTat"]
+
+demo = gr.Blocks(css=custom_css)
+
+
+def load_image(image_path):
+    image = Image.open(image_path)
+    return image
+
+
+with demo:
+    with gr.Row():
+        gr.Image("./assets/logo.png", height="200px", width="200px", scale=0.1,
+                 show_download_button=False, container=False)
+        gr.HTML(TITLE, elem_id="title")
+
+    gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")
+    with gr.Tabs(elem_classes="tab-buttons") as tabs:
+        with gr.TabItem("🏅 GTBench", elem_id="llm-benchmark-tab-table", id=0):
+            with gr.Row():
+                with gr.Column():
+                    with gr.Row():
+
+                        shown_columns = gr.CheckboxGroup(
+                            choices=[
+                                'Average'
+                            ]+games,
+                            label="Select columns to show",
+                            elem_id="column-select",
+                            interactive=True,
+                        )
+                with gr.Column(min_width=320):
+                    # with gr.Box(elem_id="box-filter"):
+                    model1_column = gr.CheckboxGroup(
+                        label="Model",
+                        choices=models,
+                        interactive=True,
+                        elem_id="filter-columns-type",
+                    )
+
+                    agent1_column = gr.CheckboxGroup(
+                        label="Agents",
+                        choices=agents,
+                        interactive=True,
+                        elem_id="filter-columns-precision",
+                    )
+
+                    model2_column = gr.CheckboxGroup(
+                        label="Opponent Model",
+                        choices=opponent_models,
+                        interactive=True,
+                        elem_id="filter-columns-type",
+                    )
+                    agent2_column = gr.CheckboxGroup(
+                        label="Opponent Agents",
+                        choices=opponent_agents,
+                        interactive=True,
+                        elem_id="filter-columns-precision",
+                    )
+                    # filter_columns_size = gr.CheckboxGroup(
+                    #     label="Model sizes (in billions of parameters)",
+                    #     choices=[f'NUMERIC_INTERVALS{i}' for i in range(0, 5)],
+                    #     value=[f'NUMERIC_INTERVALS{i}' for i in range(0, 5)],
+                    #     interactive=True,
+                    #     elem_id="filter-columns-size",
+                    # )
+
+            leaderboard_table = gr.components.Dataframe(
+                value=gtbench_raw_data,
+                elem_id="leaderboard-table",
+                interactive=False,
+                visible=True,
+                # column_widths=["2%", "33%"]
+            )
+
+            game_bench_df_for_search = gr.components.Dataframe(
+                value=gtbench_raw_data,
+                elem_id="leaderboard-table",
+                interactive=False,
+                visible=False,
+                # column_widths=["2%", "33%"]
+            )
+
+            # Dummy leaderboard for handling the case when the user uses backspace key
+            # hidden_leaderboard_table_for_search = gr.components.Dataframe(
+            #     value=[],
+            #     headers=COLS,
+            #     datatype=TYPES,
+            #     visible=False,
+            # )
+            # search_bar.submit(
+            #     update_table,
+            #     [
+            #         # hidden_leaderboard_table_for_search,
+            #         # shown_columns,
+            #         # filter_columns_type,
+            #         # filter_columns_precision,
+            #         # filter_columns_size,
+            #         # deleted_models_visibility,
+            #         # flagged_models_visibility,
+            #         # search_bar,
+            #     ],
+            #     leaderboard_table,
+            # )
+
+            # # Define a hidden component that will trigger a reload only if a query parameter has be set
+            # hidden_search_bar = gr.Textbox(value="", visible=False)
+            # hidden_search_bar.change(
+            #     update_table,
+            #     [
+            #         hidden_leaderboard_table_for_search,
+            #         shown_columns,
+            #         filter_columns_type,
+            #         filter_columns_precision,
+            #         filter_columns_size,
+            #         deleted_models_visibility,
+            #         flagged_models_visibility,
+            #         search_bar,
+            #     ],
+            #     leaderboard_table,
+            # )
+            # # Check query parameter once at startup and update search bar + hidden component
+            # demo.load(load_query, inputs=[], outputs=[search_bar, hidden_search_bar])
+
+            for selector in [shown_columns, model1_column, model2_column, agent1_column, agent2_column]:
+                selector.change(
+                    update_table,
+                    [
+                        game_bench_df_for_search,
+                        shown_columns,
+                        model1_column,
+                        model2_column,
+                        agent1_column,
+                        agent2_column
+                        # filter_columns_precision,
+                        # None,  # filter_columns_size,
+                        # None,  # deleted_models_visibility,
+                        # None,  # flagged_models_visibility,
+                        # None,  # search_bar,
+                    ],
+                    leaderboard_table,
+                    queue=True,
+                )
+
+        # with gr.TabItem("📈 Metrics through time", elem_id="llm-benchmark-tab-table", id=4):
+        #     with gr.Row():
+        #         with gr.Column():
+        #             chart = create_metric_plot_obj_1(
+        #                 dummy_data_for_plot(
+        #                     ["Metric1", "Metric2", 'Metric3']),
+        #                 ["Metric1", "Metric2", "Metric3"],
+        #                 title="Average of Top Scores and Human Baseline Over Time (from last update)",
+        #             )
+        #             gr.Plot(value=chart, min_width=500)
+
+        with gr.TabItem("📝 About", elem_id="llm-benchmark-tab-table", id=2):
+            gr.Markdown(LLM_BENCHMARKS_TEXT, elem_classes="markdown-text")
+            gr.Markdown(FAQ_TEXT, elem_classes="markdown-text")
+
+        '''
+        with gr.TabItem("🚀 Submit here! ", elem_id="llm-benchmark-tab-table", id=3):
+            with gr.Column():
+                with gr.Row():
+                    gr.Markdown(EVALUATION_QUEUE_TEXT,
+                                elem_classes="markdown-text")
+
+                with gr.Column():
+                    with gr.Accordion(
+                        f"✅ Finished Evaluations ({9})",
+                        open=False,
+                    ):
+                        with gr.Row():
+                            finished_eval_table = gr.components.Dataframe(
+                                value=None,
+                                headers=EVAL_COLS,
+                                datatype=EVAL_TYPES,
+                                row_count=5,
+                            )
+                    with gr.Accordion(
+                        f"🔄 Running Evaluation Queue ({5})",
+                        open=False,
+                    ):
+                        with gr.Row():
+                            running_eval_table = gr.components.Dataframe(
+                                value=None,
+                                headers=EVAL_COLS,
+                                datatype=EVAL_TYPES,
+                                row_count=5,
+                            )
+
+                    with gr.Accordion(
+                        f"⏳ Pending Evaluation Queue ({7})",
+                        open=False,
+                    ):
+                        with gr.Row():
+                            pending_eval_table = gr.components.Dataframe(
+                                value=None,
+                                headers=EVAL_COLS,
+                                datatype=EVAL_TYPES,
+                                row_count=5,
+                            )
+            with gr.Row():
+                gr.Markdown("# ✉️✨ Submit your Agent here!",
+                            elem_classes="markdown-text")
+
+            with gr.Row():
+                with gr.Column():
+                    model_name_textbox = gr.Textbox(label="Agent name")
+                    # revision_name_textbox = gr.Textbox(
+                    #     label="Revision commit", placeholder="main")
+                    # private = gr.Checkbox(
+                    #     False, label="Private", visible=not IS_PUBLIC)
+                    model_type = gr.Dropdown(
+                        choices=[t.to_str(" : ")
+                                 for t in ModelType if t != ModelType.Unknown],
+                        label="Agent type",
+                        multiselect=False,
+                        value=ModelType.FT.to_str(" : "),
+                        interactive=True,
+                    )
+
+                # with gr.Column():
+                #     precision = gr.Dropdown(
+                #         choices=[i.value.name for i in Precision if i !=
+                #                  Precision.Unknown],
+                #         label="Precision",
+                #         multiselect=False,
+                #         value="float16",
+                #         interactive=True,
+                #     )
+                #     weight_type = gr.Dropdown(
+                #         choices=[i.value.name for i in WeightType],
+                #         label="Weights type",
+                #         multiselect=False,
+                #         value="Original",
+                #         interactive=True,
+                #     )
+                #     base_model_name_textbox = gr.Textbox(
+                #         label="Base model (for delta or adapter weights)")
+
+            submit_button = gr.Button("Submit Eval")
+            submission_result = gr.Markdown()
+    #         submit_button.click(
+    #             add_new_eval,
+    #             [
+    #                 model_name_textbox,
+    #                 base_model_name_textbox,
+    #                 revision_name_textbox,
+    #                 precision,
+    #                 private,
+    #                 weight_type,
+    #                 model_type,
+    #             ],
+    #             submission_result,
+    #         )
+
+    '''
+    with gr.Row():
+        with gr.Accordion("📙 Citation", open=False):
+            citation_button = gr.Textbox(
+                value=CITATION_BUTTON_TEXT,
+                label=CITATION_BUTTON_LABEL,
+                lines=20,
+                elem_id="citation-button",
+                show_copy_button=True,
+            )
+
+# scheduler = BackgroundScheduler()
+# scheduler.add_job(restart_space, "interval", seconds=1800)
+# scheduler.start()
+demo.launch()
+# Both launches the space and its CI
+# configure_space_ci(
+#     demo.queue(default_concurrency_limit=40),
+#     trusted_authors=[],  # add manually trusted authors
+#     private="True",  # ephemeral spaces will have same visibility as the main space. Otherwise, set to `True` or `False` explicitly.
+#     variables={},  # We overwrite HF_HOME as tmp CI spaces will have no cache
+#     secrets=["HF_TOKEN", "H4_TOKEN"],  # which secret do I want to copy from the main space? Can be a `List[str]`."HF_TOKEN", "H4_TOKEN"
+#     hardware=None,  # "cpu-basic" by default. Otherwise set to "auto" to have same hardware as the main space or any valid string value.
+#     storage=None,  # no storage by default. Otherwise set to "auto" to have same storage as the main space or any valid string value.
+# ).launch()
+
+
+# notes: opponent model , opponent agent
+# column is games

assets/gtbench_results.csv

@@ -0,0 +1,23 @@
+Model,Agent,Opponent Model,Opponent Agent,Tic-Tac-Toe,Connect Four,Breakthrough,Liar's Dice,Blind Auction,Negotiation,Kuhn Poker,Nim,Pig,Iterated Prisoner's Dilemma,
+GPT-3.5-turbo,Prompt,GPT-3.5-turbo-1106,prompt agent,0.000,0.000,0.000,0.000,0.000,0.000,0.000,0.000,0.000,0.000,0.000
+GPT-4,Prompt,GPT-3.5-turbo-1106,prompt agent,-0.111,0.080,0.320,0.800,0.040,-0.281,0.400,0.080,-0.040,0.004,0.129
+GPT-4,CoT,GPT-3.5-turbo-1106,prompt agent,-0.022,-0.080,0.560,0.240,0.069,0.135,0.440,0.040,0.040,-0.160,0.126
+GPT-3.5-turbo,CoT,GPT-3.5-turbo-1106,prompt agent,0.277,-0.320,-0.120,0.440,0.115,-0.207,0.120,-0.040,-0.160,0.126,0.023
+GPT-3.5-turbo,SC-CoT,GPT-3.5-turbo-1106,prompt agent,0.409,-0.040,-0.160,0.520,-0.120,-0.315,-0.080,0.000,-0.080,-0.155,-0.002
+GPT-3.5-turbo,ToT,GPT-3.5-turbo-1106,prompt agent,-0.045,0.240,0.160,0.000,-0.120,0.183,0.000,0.120,-0.400,-0.191,-0.005
+Codellama-34b-instruct,Prompt,GPT-3.5-turbo-1106,prompt agent,0.333,-0.100,-0.800,-0.400,-0.250,0.216,-0.160,0.360,0.120,0.600,-0.008
+Llama-2-70b-chat,SC-CoT,GPT-3.5-turbo-1106,prompt agent,-0.469,-0.160,-0.680,0.160,-0.040,0.052,0.120,0.040,0.040,0.296,-0.064
+Codellama-34b-instruct,CoT,GPT-3.5-turbo-1106,prompt agent,0.316,-0.360,-0.760,-0.320,-0.268,0.085,0.000,0.480,-0.080,0.032,-0.088
+Llama-2-70b-chat,CoT,GPT-3.5-turbo-1106,prompt agent,-0.500,0.080,-0.800,0.265,-0.086,0.128,-0.200,0.061,-0.160,0.324,-0.089
+Mistral-7b-Orca,CoT,GPT-3.5-turbo-1106,prompt agent,-0.077,-0.120,-0.320,-0.560,0.133,0.078,0.000,0.360,-0.680,0.055,-0.113
+Codellama-34b-instruct,SC-CoT,GPT-3.5-turbo-1106,prompt agent,0.122,-0.600,-0.560,-0.280,-0.348,0.095,0.000,0.160,0.120,0.008,-0.128
+Mistral-7b-Orca,SC-CoT,GPT-3.5-turbo-1106,prompt agent,-0.200,-0.080,-0.400,-0.640,0.082,0.364,-0.040,0.440,-0.840,0.013,-0.130
+Codellama-34b-instruct,ToT,GPT-3.5-turbo-1106,prompt agent,-0.021,-0.160,-0.600,-0.520,-0.304,0.098,0.000,-0.040,-0.160,0.237,-0.147
+Llama-2-70b-chat,Prompt,GPT-3.5-turbo-1106,prompt agent,-0.366,-1.000,-0.440,-0.160,-0.075,-0.033,-0.040,0.800,-0.020,-0.712,-0.205
+Mistral-7b-Orca,ToT,GPT-3.5-turbo-1106,prompt agent,-0.179,-0.800,-0.320,-0.440,-0.047,0.299,-0.200,-0.080,-0.840,0.162,-0.245
+Mistral-7b-Orca,Prompt,GPT-3.5-turbo-1106,prompt agent,-0.429,-0.840,-0.680,-0.680,-0.069,-0.114,-0.040,-0.080,0.000,-0.182,-0.311
+GPT-4,Prompt,GPT-4,prompt agent,0.000,0.000,0.000,0.000,0.000,0.000,0.000,0.000,0.000,0.000,0.000
+Codellama-34b-instruct,Prompt,GPT-4,prompt agent,-0.064,0.720,-0.600,-0.640,-0.148,0.000,0.080,0.160,0.040,0.342,-0.011
+Codellama-34b-instruct,CoT,GPT-4,prompt agent,0.022,0.560,-1.000,-0.800,0.449,-0.078,0.080,0.200,-0.080,0.224,-0.042
+Llama-2-70b-chat,Prompt,GPT-4,prompt agent,-0.938,0.960,-0.920,-0.720,-0.250,0.000,-0.040,0.360,0.200,0.333,-0.101
+Llama-2-70b-chat,CoT,GPT-4,prompt agent,-0.286,0.200,-0.880,-0.917,-0.417,0.201,0.000,-0.026,-0.360,0.173,-0.231

dummydatagen.py

@@ -0,0 +1,159 @@
+
+from datetime import datetime, timedelta
+import numpy as np
+import pandas as pd
+import plotly.express as px
+from plotly.graph_objs import Figure
+
+# Dummy data creation
+
+
+def dummy_data_for_plot(metrics, num_days=30):
+    dates = [datetime.now() - timedelta(days=i) for i in range(num_days)]
+    data = []
+
+    for metric in metrics:
+        for date in dates:
+            model = f"Model_{metric}"
+            score = np.random.uniform(50, 55)
+            data.append([date, metric, score, model])
+
+    df = pd.DataFrame(data, columns=["date", "task", "score", "model"])
+    return df
+
+
+def create_metric_plot_obj_1(
+    df: pd.DataFrame, metrics: list[str], title: str
+) -> Figure:
+    """
+    Create a Plotly figure object with lines representing different metrics
+    and horizontal dotted lines representing human baselines.
+
+    :param df: The DataFrame containing the metric values, names, and dates.
+    :param metrics: A list of strings representing the names of the metrics
+                    to be included in the plot.
+    :param title: A string representing the title of the plot.
+    :return: A Plotly figure object with lines representing metrics and
+             horizontal dotted lines representing human baselines.
+    """
+
+    # Filter the DataFrame based on the specified metrics
+    df = df[df["task"].isin(metrics)]
+
+    # Filter the human baselines based on the specified metrics
+    # filtered_human_baselines = {k: v for k, v in HUMAN_BASELINE.items() if k in metrics}
+
+    # Create a line figure using plotly express with specified markers and custom data
+    fig = px.line(
+        df,
+        x="date",
+        y="score",
+        color="task",
+        markers=True,
+        custom_data=["task", "score", "model"],
+        title=title,
+    )
+
+    # Update hovertemplate for better hover interaction experience
+    fig.update_traces(
+        hovertemplate="<br>".join(
+            [
+                "Model Name: %{customdata[2]}",
+                "Metric Name: %{customdata[0]}",
+                "Date: %{x}",
+                "Metric Value: %{y}",
+            ]
+        )
+    )
+
+    # Update the range of the y-axis
+    fig.update_layout(yaxis_range=[0, 100])
+
+    # Create a dictionary to hold the color mapping for each metric
+    metric_color_mapping = {}
+
+    # Map each metric name to its color in the figure
+    for trace in fig.data:
+        metric_color_mapping[trace.name] = trace.line.color
+
+    # Iterate over filtered human baselines and add horizontal lines to the figure
+    # for metric, value in filtered_human_baselines.items():
+    #     color = metric_color_mapping.get(metric, "blue")  # Retrieve color from mapping; default to blue if not found
+    #     location = "top left" if metric == "HellaSwag" else "bottom left"  # Set annotation position
+    #     # Add horizontal line with matched color and positioned annotation
+    #     fig.add_hline(
+    #         y=value,
+    #         line_dash="dot",
+    #         annotation_text=f"{metric} human baseline",
+    #         annotation_position=location,
+    #         annotation_font_size=10,
+    #         annotation_font_color=color,
+    #         line_color=color,
+    #     )
+
+    return fig
+
+
+def dummydf():
+    # data = [{"Model": "gpt-35-turbo-1106",
+    #          "Agent": "prompt agent",
+    #         "Opponent Model": "gpt-4",
+    #          "Opponent Agent": "prompt agent",
+    #          'Breakthrough': 0,
+    #          'Connect Four': 0,
+    #          'Blind Auction': 0,
+    #          'Kuhn Poker': 0,
+    #          "Liar's Dice": 0,
+    #          'Negotiation': 0,
+    #          'Nim': 0,
+    #          'Pig': 0,
+    #          'Iterated Prisoners Dilemma': 0,
+    #          'Tic-Tac-Toe': 0
+    #          },
+    #         {"Model": "Llama-2-70b-chat-hf",
+    #         "Agent": "prompt agent",
+    #          "Opponent Model": "gpt-4",
+    #          "Opponent Agent": "prompt agent",
+    #          'Breakthrough': 1,
+    #          'Connect Four': 0,
+    #          'Blind Auction': 0,
+    #          'Kuhn Poker': 0,
+    #          "Liar's Dice": 0,
+    #          'Negotiation': 0,
+    #          'Nim': 0,
+    #          'Pig': 0,
+    #          'Iterated Prisoners Dilemma': 0,
+    #          'Tic-Tac-Toe': 0
+    #          },
+    #         {"Model": "gpt-35-turbo-1106",
+    #          "Agent": "ToT agent",
+    #         "Opponent Model": "gpt-4",
+    #          "Opponent Agent": "prompt agent",
+    #          'Breakthrough': 0,
+    #          'Connect Four': 0,
+    #          'Blind Auction': 0,
+    #          'Kuhn Poker': 0,
+    #          "Liar's Dice": 0,
+    #          'Negotiation': 0,
+    #          'Nim': 0,
+    #          'Pig': 0,
+    #          'Iterated Prisoners Dilemma': 0,
+    #          'Tic-Tac-Toe': 0
+    #          },
+    #         {"Model": "Llama-2-70b-chat-hf",
+    #         "Agent": "CoT agent",
+    #          "Opponent Model": "gpt-4",
+    #          "Opponent Agent": "prompt agent",
+    #          'Breakthrough': 0,
+    #          'Connect Four': 0,
+    #          'Blind Auction': 0,
+    #          'Kuhn Poker': 0,
+    #          "Liar's Dice": 0,
+    #          'Negotiation': 0,
+    #          'Nim': 0,
+    #          'Pig': 0,
+    #          'Iterated Prisoners Dilemma': 0,
+    #          'Tic-Tac-Toe': 0
+    #          }]
+    df = pd.read_csv('./assets/gtbench_results.csv')
+    return df

requirements.txt

@@ -0,0 +1,72 @@
+aiofiles==23.2.1
+altair==5.2.0
+annotated-types==0.6.0
+anyio==4.2.0
+attrs==23.2.0
+certifi==2024.2.2
+charset-normalizer==3.3.2
+click==8.1.7
+colorama==0.4.6
+contourpy==1.2.0
+cycler==0.12.1
+exceptiongroup==1.2.0
+fastapi==0.109.2
+ffmpy==0.3.1
+filelock==3.13.1
+fonttools==4.48.1
+fsspec==2024.2.0
+gradio==4.17.0
+gradio_client==0.9.0
+h11==0.14.0
+httpcore==1.0.2
+httpx==0.26.0
+huggingface-hub==0.20.3
+idna==3.6
+importlib-resources==6.1.1
+Jinja2==3.1.3
+jsonschema==4.21.1
+jsonschema-specifications==2023.12.1
+kiwisolver==1.4.5
+markdown-it-py==3.0.0
+MarkupSafe==2.1.5
+matplotlib==3.7.1
+mdurl==0.1.2
+numpy==1.24.2
+orjson==3.9.13
+packaging==23.2
+pandas==2.0.0
+pillow==10.2.0
+plotly==5.18.0
+pydantic==2.6.1
+pydantic_core==2.16.2
+pydub==0.25.1
+Pygments==2.17.2
+pyparsing==3.1.1
+python-dateutil==2.8.2
+python-multipart==0.0.7
+pytz==2024.1
+PyYAML==6.0.1
+referencing==0.33.0
+regex==2023.12.25
+requests==2.28.2
+rich==13.7.0
+rpds-py==0.17.1
+ruff==0.2.1
+safetensors==0.4.2
+semantic-version==2.10.0
+shellingham==1.5.4
+six==1.16.0
+sniffio==1.3.0
+starlette==0.36.3
+tenacity==8.2.3
+tokenizers==0.15.1
+tomlkit==0.12.0
+toolz==0.12.1
+tqdm==4.66.1
+transformers==4.36.0
+typer==0.9.0
+typing_extensions==4.9.0
+tzdata==2023.4
+urllib3==1.26.18
+uvicorn==0.27.0.post1
+websockets==11.0.3

src/display/about.py

@@ -0,0 +1,70 @@
+# from src.display.utils import ModelType
+
+TITLE = """
+<h1 id="space-title">UNLEARNCANVAS: A Stylized Image Dataset to Benchmark <br> Machine Unlearning for Diffusion Models</h1>"""
+
+INTRODUCTION_TEXT = """
+
+paper: https://arxiv.org/abs/2402.11846
+
+The rapid advancement of diffusion models (DMs) has not only transformed various real- world industries but has also introduced negative societal concerns, including the generation of harmful content, copyright disputes, and the rise of stereotypes and biases. To mitigate these issues, machine unlearning (MU) has emerged as a potential solution, demonstrating its ability to remove undesired generative capabilities of DMs in various applications. However, by examining existing MU evaluation methods, we uncover several key challenges that can result in incomplete, inaccurate, or biased evaluations for MU in DMs. 
+
+To address them, we enhance the evaluation metrics for MU, including the introduction of an often-overlooked retainability measurement for DMs post-unlearning. Additionally, we introduce UnlearnCanvas, a comprehensive high-resolution stylized image dataset that facilitates us to evaluate the unlearning of artistic painting styles in conjunction with associated image objects. 
+
+We show that this dataset plays a pivotal role in establishing a standardized and automated evaluation framework for MU techniques on DMs, featuring 7 quantitative metrics to address various aspects of unlearning effectiveness. Through extensive experiments, we benchmark 5 state-of- the-art MU methods, revealing novel insights into their pros and cons, and the underlying unlearning mechanisms. Furthermore, we demonstrate the potential of UnlearnCanvas to benchmark other generative modeling tasks, such as style transfer. The UnlearnCanvas dataset, benchmark, and the codes to reproduce all the results in this work can be found at https://github. com/OPTML-Group/UnlearnCanvas.
+
+"""
+
+LLM_BENCHMARKS_TEXT = f"""
+# Context
+
+## How it works
+
+We evaluate LLMs on 10 widely recognized game-theoretic tasks, including
+
+- <a href="https://en.wikipedia.org/wiki/Tic-tac-toe" target="_blank">  Tic-Tac-Toe</a>
+- <a href="https://en.wikipedia.org/wiki/Connect_Four" target="_blank">  Connect-4 </a>
+- <a href="https://en.wikipedia.org/wiki/Breakthrough_(board_game)" target="_blank">  Breakthrough</a>
+- <a href="https://en.wikipedia.org/wiki/Nim" target="_blank">  Nim</a>
+- <a href="https://en.wikipedia.org/wiki/First-price_sealed-bid_auction" target="_blank">  Blind Auction</a>
+- <a href="https://en.wikipedia.org/wiki/Kuhn_poker" target="_blank">  Kuhn Poker</a>
+- <a href="https://en.wikipedia.org/wiki/Liar\%27s_dice" target="_blank">  Liar's Dice</a>
+- <a href="https://arxiv.org/pdf/1706.05125.pdf" target="_blank">  Negotiation</a>
+- <a href="https://en.wikipedia.org/wiki/Pig_(dice_game)" target="_blank">  Pig</a>
+- <a href="https://en.wikipedia.org/wiki/Prisoner\%27s_dilemma" target="_blank">  Prisoner's Dilemma</a>
+
+## Metric
+We use Normalized Relative Advantage (NRA) to evaluation the performance of LLM agents. NRA(agent1, agent2) > 0 means agent1 has higher win rate/earn more rewards than the opponent agent2.
+
+Please refer to GTBench paper for more detail.
+
+## Takeaways
+- LLM agents failed in the complete-information and deterministic games
+- LLM agents are competitive in the probabilistic games
+- CodePretraining benefits game-theoretic tasks.
+- Advanced Reasoning Methods Do Not Always Help.
+
+## Contact
+Please feel free to contact Jinhao <[email protected]> and Renming <[email protected]> if you have any questions.
+
+"""
+
+FAQ_TEXT = """
+"""
+
+
+EVALUATION_QUEUE_TEXT = """
+# Evaluation for the GTBench leaderboard 
+
+"""
+
+CITATION_BUTTON_LABEL = "Copy the following snippet to cite these results"
+
+CITATION_BUTTON_TEXT = r"""
+@article{zhang2024unlearncanvas,
+  title={UnlearnCanvas: A Stylized Image Dataset to Benchmark Machine Unlearning for Diffusion Models},
+  author={Zhang, Yihua and Zhang, Yimeng and Yao, Yuguang and Jia, Jinghan and Liu, Jiancheng and Liu, Xiaoming and Liu, Sijia},
+  journal={arXiv preprint arXiv:2402.11846},
+  year={2024}
+}
+"""

src/display/css_html_js.py

@@ -0,0 +1,114 @@
+custom_css = """
+
+.markdown-text {
+    font-size: 16px !important;
+}
+
+#models-to-add-text {
+    font-size: 18px !important;
+}
+
+#citation-button span {
+    font-size: 16px !important;
+}
+
+#citation-button textarea {
+    font-size: 16px !important;
+}
+
+#citation-button > label > button {
+    margin: 6px;
+    transform: scale(1.3);
+}
+
+#leaderboard-table {
+    margin-top: 15px
+}
+
+#leaderboard-table-lite {
+    margin-top: 15px
+}
+
+#search-bar-table-box > div:first-child {
+    background: none;
+    border: none;
+}
+ 
+#search-bar {
+    padding: 0px;
+}
+
+
+
+/* Limit the width of the first AutoEvalColumn so that names don't expand too much */
+table td:first-child,
+table th:first-child {
+    max-width: 400px;
+    overflow: auto;
+    white-space: nowrap;
+}
+
+.tab-buttons button {
+    font-size: 20px;
+}
+
+#scale-logo {
+    border-style: none !important;
+    box-shadow: none;
+    display: block;
+    margin-left: auto;
+    margin-right: auto;
+    max-width: 600px;
+}
+
+#scale-logo .download {
+    display: none;
+}
+#filter_type{
+    border: 0;
+    padding-left: 0;
+    padding-top: 0;
+}
+#filter_type label {
+    display: flex;
+}
+#filter_type label > span{
+    margin-top: var(--spacing-lg);
+    margin-right: 0.5em;
+}
+#filter_type label > .wrap{
+    width: 103px;
+}
+#filter_type label > .wrap .wrap-inner{  
+    padding: 2px;
+}
+#filter_type label > .wrap .wrap-inner input{
+    width: 1px
+}
+#filter-columns-type{
+    border:0;
+    padding:0.5;
+}
+#filter-columns-size{
+    border:0;
+    padding:0.5;
+}
+#box-filter > .form{
+    border: 0
+}
+#title{
+    margin-top: 110px
+    text-align: left;
+    display: flex;
+    justify-content: flex-start;
+    align-items: center;
+}
+"""
+
+get_window_url_params = """
+    function(url_params) {
+        const params = new URLSearchParams(window.location.search);
+        url_params = Object.fromEntries(params);
+        return url_params;
+    }
+    """

src/display/formatting.py

@@ -0,0 +1,40 @@
+import os
+from datetime import datetime, timezone
+
+from huggingface_hub import HfApi
+from huggingface_hub.hf_api import ModelInfo
+
+
+API = HfApi()
+
+def model_hyperlink(link, model_name):
+    return f'<a target="_blank" href="{link}" style="color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;">{model_name}</a>'
+
+
+def make_clickable_model(model_name):
+    link = f"https://huggingface.co/{model_name}"
+
+    details_model_name = model_name.replace("/", "__")
+    details_link = f"https://huggingface.co/datasets/open-llm-leaderboard/details_{details_model_name}"
+
+    return model_hyperlink(link, model_name) + "  " + model_hyperlink(details_link, "📑")
+
+
+def styled_error(error):
+    return f"<p style='color: red; font-size: 20px; text-align: center;'>{error}</p>"
+
+
+def styled_warning(warn):
+    return f"<p style='color: orange; font-size: 20px; text-align: center;'>{warn}</p>"
+
+
+def styled_message(message):
+    return f"<p style='color: green; font-size: 20px; text-align: center;'>{message}</p>"
+
+
+def has_no_nan_values(df, columns):
+    return df[columns].notna().all(axis=1)
+
+
+def has_nan_values(df, columns):
+    return df[columns].isna().any(axis=1)

src/display/utils.py

@@ -0,0 +1,190 @@
+from dataclasses import dataclass, make_dataclass
+from enum import Enum
+
+import pandas as pd
+
+def fields(raw_class):
+    return [v for k, v in raw_class.__dict__.items() if k[:2] != "__" and k[-2:] != "__"]
+
+
+@dataclass
+class Task:
+    benchmark: str
+    metric: str
+    col_name: str
+
+class Tasks(Enum):
+    arc = Task("arc:challenge", "acc_norm", "ARC")
+    hellaswag = Task("hellaswag", "acc_norm", "HellaSwag")
+    mmlu = Task("hendrycksTest", "acc", "MMLU")
+    truthfulqa = Task("truthfulqa:mc", "mc2", "TruthfulQA")
+    winogrande = Task("winogrande", "acc", "Winogrande")
+    gsm8k = Task("gsm8k", "acc", "GSM8K")
+
+# These classes are for user facing column names,
+# to avoid having to change them all around the code
+# when a modif is needed
+@dataclass
+class ColumnContent:
+    name: str
+    type: str
+    displayed_by_default: bool
+    hidden: bool = False
+    never_hidden: bool = False
+    dummy: bool = False
+
+auto_eval_column_dict = []
+# Init
+auto_eval_column_dict.append(["model_type_symbol", ColumnContent, ColumnContent("T", "str", True, never_hidden=True)])
+auto_eval_column_dict.append(["model", ColumnContent, ColumnContent("Model", "markdown", True, never_hidden=True)])
+#Scores
+auto_eval_column_dict.append(["average", ColumnContent, ColumnContent("Average ⬆️", "number", True)])
+for task in Tasks:
+    auto_eval_column_dict.append([task.name, ColumnContent, ColumnContent(task.value.col_name, "number", True)])
+# Model information
+auto_eval_column_dict.append(["model_type", ColumnContent, ColumnContent("Type", "str", False)])
+auto_eval_column_dict.append(["architecture", ColumnContent, ColumnContent("Architecture", "str", False)])
+auto_eval_column_dict.append(["weight_type", ColumnContent, ColumnContent("Weight type", "str", False, True)])
+auto_eval_column_dict.append(["precision", ColumnContent, ColumnContent("Precision", "str", False)])
+auto_eval_column_dict.append(["merge", ColumnContent, ColumnContent("Merged", "bool", False)])
+auto_eval_column_dict.append(["license", ColumnContent, ColumnContent("Hub License", "str", False)])
+auto_eval_column_dict.append(["params", ColumnContent, ColumnContent("#Params (B)", "number", False)])
+auto_eval_column_dict.append(["likes", ColumnContent, ColumnContent("Hub ❤️", "number", False)])
+auto_eval_column_dict.append(["still_on_hub", ColumnContent, ColumnContent("Available on the hub", "bool", False)])
+auto_eval_column_dict.append(["revision", ColumnContent, ColumnContent("Model sha", "str", False, False)])
+auto_eval_column_dict.append(["flagged", ColumnContent, ColumnContent("Flagged", "bool", False, False)])
+# Dummy column for the search bar (hidden by the custom CSS)
+auto_eval_column_dict.append(["dummy", ColumnContent, ColumnContent("model_name_for_query", "str", False, dummy=True)])
+
+# We use make dataclass to dynamically fill the scores from Tasks
+AutoEvalColumn = make_dataclass("AutoEvalColumn", auto_eval_column_dict, frozen=True)
+
+@dataclass(frozen=True)
+class EvalQueueColumn:  # Queue column
+    model = ColumnContent("model", "markdown", True)
+    revision = ColumnContent("revision", "str", True)
+    private = ColumnContent("private", "bool", True)
+    precision = ColumnContent("precision", "str", True)
+    weight_type = ColumnContent("weight_type", "str", "Original")
+    status = ColumnContent("status", "str", True)
+
+
+baseline_row = {
+    AutoEvalColumn.model.name: "<p>Baseline</p>",
+    AutoEvalColumn.revision.name: "N/A",
+    AutoEvalColumn.precision.name: None,
+    AutoEvalColumn.merge.name: False,
+    AutoEvalColumn.average.name: 31.0,
+    AutoEvalColumn.arc.name: 25.0,
+    AutoEvalColumn.hellaswag.name: 25.0,
+    AutoEvalColumn.mmlu.name: 25.0,
+    AutoEvalColumn.truthfulqa.name: 25.0,
+    AutoEvalColumn.winogrande.name: 50.0,
+    AutoEvalColumn.gsm8k.name: 0.21,
+    AutoEvalColumn.dummy.name: "baseline",
+    AutoEvalColumn.model_type.name: "",
+    AutoEvalColumn.flagged.name: False,
+}
+
+# Average ⬆️ human baseline is 0.897 (source: averaging human baselines below)
+# ARC human baseline is 0.80 (source: https://lab42.global/arc/)
+# HellaSwag human baseline is 0.95 (source: https://deepgram.com/learn/hellaswag-llm-benchmark-guide)
+# MMLU human baseline is 0.898 (source: https://openreview.net/forum?id=d7KBjmI3GmQ)
+# TruthfulQA human baseline is 0.94(source: https://arxiv.org/pdf/2109.07958.pdf)
+# Winogrande: https://leaderboard.allenai.org/winogrande/submissions/public
+# GSM8K: paper
+# Define the human baselines
+human_baseline_row = {
+    AutoEvalColumn.model.name: "<p>Human performance</p>",
+    AutoEvalColumn.revision.name: "N/A",
+    AutoEvalColumn.precision.name: None,
+    AutoEvalColumn.average.name: 92.75,
+    AutoEvalColumn.merge.name: False,
+    AutoEvalColumn.arc.name: 80.0,
+    AutoEvalColumn.hellaswag.name: 95.0,
+    AutoEvalColumn.mmlu.name: 89.8,
+    AutoEvalColumn.truthfulqa.name: 94.0,
+    AutoEvalColumn.winogrande.name: 94.0,
+    AutoEvalColumn.gsm8k.name: 100,
+    AutoEvalColumn.dummy.name: "human_baseline",
+    AutoEvalColumn.model_type.name: "",
+}
+
+@dataclass
+class ModelDetails:
+    name: str
+    symbol: str = "" # emoji, only for the model type
+
+
+class ModelType(Enum):
+    PT = ModelDetails(name="pretrained", symbol="🟢")
+    FT = ModelDetails(name="fine-tuned", symbol="🔶")
+    IFT = ModelDetails(name="instruction-tuned", symbol="⭕")
+    RL = ModelDetails(name="RL-tuned", symbol="🟦")
+    Unknown = ModelDetails(name="", symbol="?")
+
+    def to_str(self, separator=" "):
+        return f"{self.value.symbol}{separator}{self.value.name}"
+
+    @staticmethod
+    def from_str(type):
+        if "fine-tuned" in type or "🔶" in type:
+            return ModelType.FT
+        if "pretrained" in type or "🟢" in type:
+            return ModelType.PT
+        if "RL-tuned" in type or "🟦" in type:
+            return ModelType.RL
+        if "instruction-tuned" in type or "⭕" in type:
+            return ModelType.IFT
+        return ModelType.Unknown
+
+class WeightType(Enum):
+    Adapter = ModelDetails("Adapter")
+    Original = ModelDetails("Original")
+    Delta = ModelDetails("Delta")
+
+class Precision(Enum):
+    float16 = ModelDetails("float16")
+    bfloat16 = ModelDetails("bfloat16")
+    qt_8bit = ModelDetails("8bit")
+    qt_4bit = ModelDetails("4bit")
+    qt_GPTQ = ModelDetails("GPTQ")
+    Unknown = ModelDetails("?")
+
+    def from_str(precision):
+        if precision in ["torch.float16", "float16"]:
+            return Precision.float16
+        if precision in ["torch.bfloat16", "bfloat16"]:
+            return Precision.bfloat16
+        if precision in ["8bit"]:
+            return Precision.qt_8bit
+        if precision in ["4bit"]:
+            return Precision.qt_4bit
+        if precision in ["GPTQ", "None"]:
+            return Precision.qt_GPTQ
+        return Precision.Unknown
+        
+
+
+
+# Column selection
+COLS = [c.name for c in fields(AutoEvalColumn) if not c.hidden]
+TYPES = [c.type for c in fields(AutoEvalColumn) if not c.hidden]
+COLS_LITE = [c.name for c in fields(AutoEvalColumn) if c.displayed_by_default and not c.hidden]
+TYPES_LITE = [c.type for c in fields(AutoEvalColumn) if c.displayed_by_default and not c.hidden]
+
+EVAL_COLS = [c.name for c in fields(EvalQueueColumn)]
+EVAL_TYPES = [c.type for c in fields(EvalQueueColumn)]
+
+BENCHMARK_COLS = [t.value.col_name for t in Tasks]
+
+NUMERIC_INTERVALS = {
+    "?": pd.Interval(-1, 0, closed="right"),
+    "~1.5": pd.Interval(0, 2, closed="right"),
+    "~3": pd.Interval(2, 4, closed="right"),
+    "~7": pd.Interval(4, 9, closed="right"),
+    "~13": pd.Interval(9, 20, closed="right"),
+    "~35": pd.Interval(20, 45, closed="right"),
+    "~60": pd.Interval(45, 70, closed="right"),
+    "70+": pd.Interval(70, 10000, closed="right"),
+}

src/envs.py

@@ -0,0 +1,32 @@
+import os
+
+from huggingface_hub import HfApi
+
+# clone / pull the lmeh eval data
+H4_TOKEN = os.environ.get("H4_TOKEN", None)
+
+REPO_ID = "HuggingFaceH4/open_llm_leaderboard"
+QUEUE_REPO = "open-llm-leaderboard/requests"
+RESULTS_REPO = "open-llm-leaderboard/results"
+
+PRIVATE_QUEUE_REPO = "open-llm-leaderboard/private-requests"
+PRIVATE_RESULTS_REPO = "open-llm-leaderboard/private-results"
+
+IS_PUBLIC = bool(os.environ.get("IS_PUBLIC", True))
+
+CACHE_PATH=os.getenv("HF_HOME", ".")
+
+EVAL_REQUESTS_PATH = os.path.join(CACHE_PATH, "eval-queue")
+EVAL_RESULTS_PATH = os.path.join(CACHE_PATH, "eval-results")
+
+EVAL_REQUESTS_PATH_PRIVATE = "eval-queue-private"
+EVAL_RESULTS_PATH_PRIVATE = "eval-results-private"
+
+PATH_TO_COLLECTION = "open-llm-leaderboard/llm-leaderboard-best-models-652d6c7965a4619fb5c27a03"
+
+# Rate limit variables
+RATE_LIMIT_PERIOD = 7
+RATE_LIMIT_QUOTA = 5
+HAS_HIGHER_RATE_LIMIT = ["TheBloke"]
+
+API = HfApi(token=H4_TOKEN)

src/leaderboard/filter_models.py

@@ -0,0 +1,81 @@
+from src.display.formatting import model_hyperlink
+from src.display.utils import AutoEvalColumn
+
+# Models which have been flagged by users as being problematic for a reason or another
+# (Model name to forum discussion link)
+FLAGGED_MODELS = {
+    "Voicelab/trurl-2-13b": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/202",
+    "deepnight-research/llama-2-70B-inst": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/207",
+    "Aspik101/trurl-2-13b-pl-instruct_unload": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/213",
+    "Fredithefish/ReasonixPajama-3B-HF": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/236",
+    "TigerResearch/tigerbot-7b-sft-v1": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/237",
+    "gaodrew/gaodrew-gorgonzola-13b": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/215",
+    "AIDC-ai-business/Marcoroni-70B": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/287",
+    "AIDC-ai-business/Marcoroni-13B": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/287",
+    "AIDC-ai-business/Marcoroni-7B": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/287",
+    "fblgit/una-xaberius-34b-v1beta": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/444",
+    "jan-hq/trinity-v1": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/474",
+    "rwitz2/go-bruins-v2.1.1": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/474",
+    "rwitz2/go-bruins-v2.1": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/474",
+    "GreenNode/GreenNodeLM-v3olet-7B": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/474",
+    "GreenNode/GreenNodeLM-7B-v4leo": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/474",
+    "GreenNode/LeoScorpius-GreenNode-7B-v1": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/474",
+    "viethq188/LeoScorpius-7B-Chat-DPO": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/474",
+    "GreenNode/GreenNodeLM-7B-v2leo": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/474",
+    "janai-hq/trinity-v1": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/474",
+    "ignos/LeoScorpius-GreenNode-Alpaca-7B-v1": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/474",
+    "fblgit/una-cybertron-7b-v3-OMA": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/474",
+    "mncai/mistral-7b-dpo-merge-v1.1": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/474",
+    "mncai/mistral-7b-dpo-v6": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/474",
+    "Toten5/LeoScorpius-GreenNode-7B-v1": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/474",
+    "GreenNode/GreenNodeLM-7B-v1olet":  "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/474",
+    "quantumaikr/quantum-dpo-v0.1": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/474",
+    "quantumaikr/quantum-v0.01": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/474",
+    "quantumaikr/quantum-trinity-v0.1": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/474",
+    "mncai/mistral-7b-dpo-v5": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/474",
+    "cookinai/BruinHermes": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/474",
+    "jan-ai/Pandora-10.7B-v1": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/474",
+    "v1olet/v1olet_marcoroni-go-bruins-merge-7B": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/474",
+    "v1olet/v1olet_merged_dpo_7B_v3": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/474",
+    "rwitz2/pee": "https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/474",
+}
+
+# Models which have been requested by orgs to not be submitted on the leaderboard
+DO_NOT_SUBMIT_MODELS = [
+    "Voicelab/trurl-2-13b",  # trained on MMLU
+    "TigerResearch/tigerbot-70b-chat", # per authors request
+    "TigerResearch/tigerbot-70b-chat-v2", # per authors request
+    "TigerResearch/tigerbot-70b-chat-v4-4k", # per authors request
+]
+
+
+def flag_models(leaderboard_data: list[dict]):
+    for model_data in leaderboard_data:
+        if model_data["model_name_for_query"] in FLAGGED_MODELS:
+            issue_num = FLAGGED_MODELS[model_data["model_name_for_query"]].split("/")[-1]
+            issue_link = model_hyperlink(
+                FLAGGED_MODELS[model_data["model_name_for_query"]],
+                f"See discussion #{issue_num}",
+            )
+            model_data[
+                AutoEvalColumn.model.name
+            ] = f"{model_data[AutoEvalColumn.model.name]} has been flagged! {issue_link}"
+            model_data[AutoEvalColumn.flagged.name] = True
+        else:
+            model_data[AutoEvalColumn.flagged.name] = False
+
+
+def remove_forbidden_models(leaderboard_data: list[dict]):
+    indices_to_remove = []
+    for ix, model in enumerate(leaderboard_data):
+        if model["model_name_for_query"] in DO_NOT_SUBMIT_MODELS:
+            indices_to_remove.append(ix)
+
+    for ix in reversed(indices_to_remove):
+        leaderboard_data.pop(ix)
+    return leaderboard_data
+
+
+def filter_models(leaderboard_data: list[dict]):
+    leaderboard_data = remove_forbidden_models(leaderboard_data)
+    flag_models(leaderboard_data)

src/leaderboard/read_evals.py

@@ -0,0 +1,222 @@
+import glob
+import json
+import math
+import os
+from dataclasses import dataclass
+
+import dateutil
+import numpy as np
+
+from huggingface_hub import ModelCard
+
+from src.display.formatting import make_clickable_model
+from src.display.utils import AutoEvalColumn, ModelType, Tasks, Precision, WeightType
+# from src.submission.check_validity import is_model_on_hub
+
+
+@dataclass
+class EvalResult:
+    # Also see src.display.utils.AutoEvalColumn for what will be displayed.
+    eval_name: str # org_model_precision (uid)
+    full_model: str # org/model (path on hub)
+    org: str 
+    model: str
+    revision: str # commit hash, "" if main
+    results: dict
+    precision: Precision = Precision.Unknown
+    model_type: ModelType = ModelType.Unknown # Pretrained, fine tuned, ...
+    weight_type: WeightType = WeightType.Original # Original or Adapter
+    architecture: str = "Unknown" # From config file
+    license: str = "?"
+    likes: int = 0
+    num_params: int = 0
+    date: str = "" # submission date of request file
+    still_on_hub: bool = False
+    merge: bool = False
+
+    @classmethod
+    def init_from_json_file(self, json_filepath):
+        """Inits the result from the specific model result file"""
+        with open(json_filepath) as fp:
+            data = json.load(fp)
+
+        # We manage the legacy config format
+        config = data.get("config", data.get("config_general", None))
+
+        # Precision
+        precision = Precision.from_str(config.get("model_dtype"))
+
+        # Get model and org
+        org_and_model = config.get("model_name", config.get("model_args", None))
+        org_and_model = org_and_model.split("/", 1)
+
+        if len(org_and_model) == 1:
+            org = None
+            model = org_and_model[0]
+            result_key = f"{model}_{precision.value.name}"
+        else:
+            org = org_and_model[0]
+            model = org_and_model[1]
+            result_key = f"{org}_{model}_{precision.value.name}"
+        full_model = "/".join(org_and_model)
+
+        try:
+            merge = any(t in ["merge", "mergedlm"] for t in ModelCard.load(full_model).data.tags)
+        except Exception:
+            merge = False
+
+        still_on_hub, error, model_config = is_model_on_hub(
+            full_model, config.get("model_sha", "main"), trust_remote_code=True, test_tokenizer=False
+        )
+        architecture = "?"
+        if model_config is not None:
+            architectures = getattr(model_config, "architectures", None)
+            if architectures:
+                architecture = ";".join(architectures)
+
+        # Extract results available in this file (some results are split in several files)
+        results = {}
+        for task in Tasks:
+            task = task.value
+            # We skip old mmlu entries
+            wrong_mmlu_version = False
+            if task.benchmark == "hendrycksTest":
+                for mmlu_k in ["harness|hendrycksTest-abstract_algebra|5", "hendrycksTest-abstract_algebra"]:
+                    if mmlu_k in data["versions"] and data["versions"][mmlu_k] == 0:
+                        wrong_mmlu_version = True
+
+            if wrong_mmlu_version:
+                continue
+
+            # Some truthfulQA values are NaNs
+            if task.benchmark == "truthfulqa:mc" and "harness|truthfulqa:mc|0" in data["results"]:
+                if math.isnan(float(data["results"]["harness|truthfulqa:mc|0"][task.metric])):
+                    results[task.benchmark] = 0.0
+                    continue
+
+            # We average all scores of a given metric (mostly for mmlu)
+            accs = np.array([v.get(task.metric, None) for k, v in data["results"].items() if task.benchmark in k])
+            if accs.size == 0 or any([acc is None for acc in accs]):
+                continue
+
+            mean_acc = np.mean(accs) * 100.0
+            results[task.benchmark] = mean_acc
+
+        return self(
+            eval_name=result_key,
+            full_model=full_model,
+            org=org,
+            model=model,
+            results=results,
+            precision=precision,  
+            revision= config.get("model_sha", ""),
+            still_on_hub=still_on_hub,
+            architecture=architecture,
+            merge=merge
+        )
+
+    def update_with_request_file(self, requests_path):
+        """Finds the relevant request file for the current model and updates info with it"""
+        request_file = get_request_file_for_model(requests_path, self.full_model, self.precision.value.name)
+
+        try:
+            with open(request_file, "r") as f:
+                request = json.load(f)
+            self.model_type = ModelType.from_str(request.get("model_type", ""))
+            self.weight_type = WeightType[request.get("weight_type", "Original")]
+            self.license = request.get("license", "?")
+            self.likes = request.get("likes", 0)
+            self.num_params = request.get("params", 0)
+            self.date = request.get("submitted_time", "")
+        except Exception:
+            print(f"Could not find request file for {self.org}/{self.model}")
+
+    def to_dict(self):
+        """Converts the Eval Result to a dict compatible with our dataframe display"""
+        average = sum([v for v in self.results.values() if v is not None]) / len(Tasks)
+        data_dict = {
+            "eval_name": self.eval_name,  # not a column, just a save name,
+            AutoEvalColumn.precision.name: self.precision.value.name,
+            AutoEvalColumn.model_type.name: self.model_type.value.name,
+            AutoEvalColumn.merge.name: self.merge,
+            AutoEvalColumn.model_type_symbol.name: self.model_type.value.symbol,
+            AutoEvalColumn.weight_type.name: self.weight_type.value.name,
+            AutoEvalColumn.architecture.name: self.architecture,
+            AutoEvalColumn.model.name: make_clickable_model(self.full_model),
+            AutoEvalColumn.dummy.name: self.full_model,
+            AutoEvalColumn.revision.name: self.revision,
+            AutoEvalColumn.average.name: average,
+            AutoEvalColumn.license.name: self.license,
+            AutoEvalColumn.likes.name: self.likes,
+            AutoEvalColumn.params.name: self.num_params,
+            AutoEvalColumn.still_on_hub.name: self.still_on_hub,
+        }
+
+        for task in Tasks:
+            data_dict[task.value.col_name] = self.results[task.value.benchmark]
+
+        return data_dict
+
+
+def get_request_file_for_model(requests_path, model_name, precision):
+    """Selects the correct request file for a given model. Only keeps runs tagged as FINISHED"""
+    request_files = os.path.join(
+        requests_path,
+        f"{model_name}_eval_request_*.json",
+    )
+    request_files = glob.glob(request_files)
+
+    # Select correct request file (precision)
+    request_file = ""
+    request_files = sorted(request_files, reverse=True)
+    for tmp_request_file in request_files:
+        with open(tmp_request_file, "r") as f:
+            req_content = json.load(f)
+            if (
+                req_content["status"] in ["FINISHED"]
+                and req_content["precision"] == precision.split(".")[-1]
+            ):
+                request_file = tmp_request_file
+    return request_file
+
+
+def get_raw_eval_results(results_path: str, requests_path: str) -> list[EvalResult]:
+    """From the path of the results folder root, extract all needed info for results"""
+    model_result_filepaths = []
+
+    for root, _, files in os.walk(results_path):
+        # We should only have json files in model results
+        if len(files) == 0 or any([not f.endswith(".json") for f in files]):
+            continue
+
+        # Sort the files by date
+        try:
+            files.sort(key=lambda x: x.removesuffix(".json").removeprefix("results_")[:-7])
+        except dateutil.parser._parser.ParserError:
+            files = [files[-1]]
+
+        for file in files:
+            model_result_filepaths.append(os.path.join(root, file))
+
+    eval_results = {}
+    for model_result_filepath in model_result_filepaths:
+        # Creation of result
+        eval_result = EvalResult.init_from_json_file(model_result_filepath)
+        eval_result.update_with_request_file(requests_path)
+
+        # Store results of same eval together
+        eval_name = eval_result.eval_name
+        if eval_name in eval_results.keys():
+            eval_results[eval_name].results.update({k: v for k, v in eval_result.results.items() if v is not None})
+        else:
+            eval_results[eval_name] = eval_result
+
+    results = []
+    for v in eval_results.values():
+        try:
+            v.to_dict() # we test if the dict version is complete
+            results.append(v)
+        except KeyError:  # not all eval values present
+            continue
+
+    return results

src/populate.py

@@ -0,0 +1,59 @@
+import json
+import os
+
+import pandas as pd
+
+from src.display.formatting import has_no_nan_values, make_clickable_model
+from src.display.utils import AutoEvalColumn, EvalQueueColumn, baseline_row
+from src.leaderboard.filter_models import filter_models
+from src.leaderboard.read_evals import get_raw_eval_results
+
+
+def get_leaderboard_df(results_path: str, requests_path: str, cols: list, benchmark_cols: list) -> pd.DataFrame:
+    raw_data = get_raw_eval_results(results_path, requests_path)
+    all_data_json = [v.to_dict() for v in raw_data]
+    all_data_json.append(baseline_row)
+    filter_models(all_data_json)
+
+    df = pd.DataFrame.from_records(all_data_json)
+    df = df.sort_values(by=[AutoEvalColumn.average.name], ascending=False)
+    df = df[cols].round(decimals=2)
+
+    # filter out if any of the benchmarks have not been produced
+    df = df[has_no_nan_values(df, benchmark_cols)]
+    return raw_data, df
+
+
+def get_evaluation_queue_df(save_path: str, cols: list) -> list[pd.DataFrame]:
+    entries = [entry for entry in os.listdir(save_path) if not entry.startswith(".")]
+    all_evals = []
+
+    for entry in entries:
+        if ".json" in entry:
+            file_path = os.path.join(save_path, entry)
+            with open(file_path) as fp:
+                data = json.load(fp)
+
+            data[EvalQueueColumn.model.name] = make_clickable_model(data["model"])
+            data[EvalQueueColumn.revision.name] = data.get("revision", "main")
+
+            all_evals.append(data)
+        elif ".md" not in entry:
+            # this is a folder
+            sub_entries = [e for e in os.listdir(f"{save_path}/{entry}") if not e.startswith(".")]
+            for sub_entry in sub_entries:
+                file_path = os.path.join(save_path, entry, sub_entry)
+                with open(file_path) as fp:
+                    data = json.load(fp)
+
+                data[EvalQueueColumn.model.name] = make_clickable_model(data["model"])
+                data[EvalQueueColumn.revision.name] = data.get("revision", "main")
+                all_evals.append(data)
+
+    pending_list = [e for e in all_evals if e["status"] in ["PENDING", "RERUN"]]
+    running_list = [e for e in all_evals if e["status"] == "RUNNING"]
+    finished_list = [e for e in all_evals if e["status"].startswith("FINISHED") or e["status"] == "PENDING_NEW_EVAL"]
+    df_pending = pd.DataFrame.from_records(pending_list, columns=cols)
+    df_running = pd.DataFrame.from_records(running_list, columns=cols)
+    df_finished = pd.DataFrame.from_records(finished_list, columns=cols)
+    return df_finished[cols], df_running[cols], df_pending[cols]

src/submission/check_validity.py

@@ -0,0 +1,130 @@
+import json
+import os
+import re
+from collections import defaultdict
+from datetime import datetime, timedelta, timezone
+
+import huggingface_hub
+from huggingface_hub import ModelCard
+from huggingface_hub.hf_api import ModelInfo
+# from transformers import AutoConfig, AutoTokenizer
+# from transformers.models.auto.tokenization_auto import tokenizer_class_from_name, get_tokenizer_config
+
+from src.envs import HAS_HIGHER_RATE_LIMIT
+
+
+# ht to @Wauplin, thank you for the snippet!
+# See https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard/discussions/317
+def check_model_card(repo_id: str) -> tuple[bool, str]:
+    # Returns operation status, and error message
+    try:
+        card = ModelCard.load(repo_id)
+    except huggingface_hub.utils.EntryNotFoundError:
+        return False, "Please add a model card to your model to explain how you trained/fine-tuned it."
+
+    # Enforce license metadata
+    if card.data.license is None:
+        if not ("license_name" in card.data and "license_link" in card.data):
+            return False, (
+                "License not found. Please add a license to your model card using the `license` metadata or a"
+                " `license_name`/`license_link` pair."
+            )
+
+    # Enforce card content
+    if len(card.text) < 200:
+        return False, "Please add a description to your model card, it is too short."
+
+    return True, ""
+
+#
+# def is_model_on_hub(model_name: str, revision: str, token: str = None, trust_remote_code=False, test_tokenizer=False) -> tuple[bool, str]:
+#     try:
+#         config = AutoConfig.from_pretrained(model_name, revision=revision, trust_remote_code=trust_remote_code, token=token)
+#         if test_tokenizer:
+#             try:
+#                 tk = AutoTokenizer.from_pretrained(model_name, revision=revision, trust_remote_code=trust_remote_code, token=token)
+#             except ValueError as e:
+#                 return (
+#                     False,
+#                     f"uses a tokenizer which is not in a transformers release: {e}",
+#                     None
+#                 )
+#             except Exception as e:
+#                 return (False, "'s tokenizer cannot be loaded. Is your tokenizer class in a stable transformers release, and correctly configured?", None)
+#         return True, None, config
+#
+#     except ValueError:
+#         return (
+#             False,
+#             "needs to be launched with `trust_remote_code=True`. For safety reason, we do not allow these models to be automatically submitted to the leaderboard.",
+#             None
+#         )
+#
+#     except Exception as e:
+#         return False, "was not found on hub!", None
+
+
+def get_model_size(model_info: ModelInfo, precision: str):
+    size_pattern = size_pattern = re.compile(r"(\d\.)?\d+(b|m)")
+    try:
+        model_size = round(model_info.safetensors["total"] / 1e9, 3)
+    except (AttributeError, TypeError ):
+        try:
+            size_match = re.search(size_pattern, model_info.modelId.lower())
+            model_size = size_match.group(0)
+            model_size = round(float(model_size[:-1]) if model_size[-1] == "b" else float(model_size[:-1]) / 1e3, 3)
+        except AttributeError:
+            return 0  # Unknown model sizes are indicated as 0, see NUMERIC_INTERVALS in app.py
+
+    size_factor = 8 if (precision == "GPTQ" or "gptq" in model_info.modelId.lower()) else 1
+    model_size = size_factor * model_size
+    return model_size
+
+def get_model_arch(model_info: ModelInfo):
+    return model_info.config.get("architectures", "Unknown")
+
+def user_submission_permission(org_or_user, users_to_submission_dates, rate_limit_period, rate_limit_quota):
+    if org_or_user not in users_to_submission_dates:
+        return True, ""
+    submission_dates = sorted(users_to_submission_dates[org_or_user])
+
+    time_limit = (datetime.now(timezone.utc) - timedelta(days=rate_limit_period)).strftime("%Y-%m-%dT%H:%M:%SZ")
+    submissions_after_timelimit = [d for d in submission_dates if d > time_limit]
+
+    num_models_submitted_in_period = len(submissions_after_timelimit)
+    if org_or_user in HAS_HIGHER_RATE_LIMIT:
+        rate_limit_quota = 2 * rate_limit_quota
+
+    if num_models_submitted_in_period > rate_limit_quota:
+        error_msg = f"Organisation or user `{org_or_user}`"
+        error_msg += f"already has {num_models_submitted_in_period} model requests submitted to the leaderboard "
+        error_msg += f"in the last {rate_limit_period} days.\n"
+        error_msg += (
+            "Please wait a couple of days before resubmitting, so that everybody can enjoy using the leaderboard 🤗"
+        )
+        return False, error_msg
+    return True, ""
+
+
+def already_submitted_models(requested_models_dir: str) -> set[str]:
+    depth = 1
+    file_names = []
+    users_to_submission_dates = defaultdict(list)
+
+    for root, _, files in os.walk(requested_models_dir):
+        current_depth = root.count(os.sep) - requested_models_dir.count(os.sep)
+        if current_depth == depth:
+            for file in files:
+                if not file.endswith(".json"):
+                    continue
+                with open(os.path.join(root, file), "r") as f:
+                    info = json.load(f)
+                    file_names.append(f"{info['model']}_{info['revision']}_{info['precision']}")
+
+                    # Select organisation
+                    if info["model"].count("/") == 0 or "submitted_time" not in info:
+                        continue
+                    organisation, _ = info["model"].split("/")
+                    users_to_submission_dates[organisation].append(info["submitted_time"])
+
+    return set(file_names), users_to_submission_dates

src/submission/submit.py

@@ -0,0 +1,134 @@
+import json
+import os
+from datetime import datetime, timezone
+
+from src.display.formatting import styled_error, styled_message, styled_warning
+from src.envs import API, EVAL_REQUESTS_PATH, H4_TOKEN, QUEUE_REPO, RATE_LIMIT_PERIOD, RATE_LIMIT_QUOTA
+from src.leaderboard.filter_models import DO_NOT_SUBMIT_MODELS
+from src.submission.check_validity import (
+    already_submitted_models,
+    check_model_card,
+    get_model_size,
+    is_model_on_hub,
+    user_submission_permission,
+)
+
+REQUESTED_MODELS = None
+USERS_TO_SUBMISSION_DATES = None
+
+def add_new_eval(
+    model: str,
+    base_model: str,
+    revision: str,
+    precision: str,
+    private: bool,
+    weight_type: str,
+    model_type: str,
+):
+    global REQUESTED_MODELS
+    global USERS_TO_SUBMISSION_DATES
+    if not REQUESTED_MODELS:
+        REQUESTED_MODELS, USERS_TO_SUBMISSION_DATES = already_submitted_models(EVAL_REQUESTS_PATH)
+
+    user_name = ""
+    model_path = model
+    if "/" in model:
+        user_name = model.split("/")[0]
+        model_path = model.split("/")[1]
+
+    precision = precision.split(" ")[0]
+    current_time = datetime.now(timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ")
+
+    if model_type is None or model_type == "":
+        return styled_error("Please select a model type.")
+
+    # Is the user rate limited?
+    if user_name != "":
+        user_can_submit, error_msg = user_submission_permission(
+            user_name, USERS_TO_SUBMISSION_DATES, RATE_LIMIT_PERIOD, RATE_LIMIT_QUOTA
+        )
+        if not user_can_submit:
+            return styled_error(error_msg)
+
+    # Did the model authors forbid its submission to the leaderboard?
+    if model in DO_NOT_SUBMIT_MODELS or base_model in DO_NOT_SUBMIT_MODELS:
+        return styled_warning("Model authors have requested that their model be not submitted on the leaderboard.")
+
+    # Does the model actually exist?
+    if revision == "":
+        revision = "main"
+
+    # Is the model on the hub?
+    if weight_type in ["Delta", "Adapter"]:
+        base_model_on_hub, error, _ = is_model_on_hub(model_name=base_model, revision=revision, token=H4_TOKEN, test_tokenizer=True)
+        if not base_model_on_hub:
+            return styled_error(f'Base model "{base_model}" {error}')
+
+    if not weight_type == "Adapter":
+        model_on_hub, error, _ = is_model_on_hub(model_name=model, revision=revision, test_tokenizer=True)
+        if not model_on_hub:
+            return styled_error(f'Model "{model}" {error}')
+
+    # Is the model info correctly filled?
+    try:
+        model_info = API.model_info(repo_id=model, revision=revision)
+    except Exception:
+        return styled_error("Could not get your model information. Please fill it up properly.")
+
+    model_size = get_model_size(model_info=model_info, precision=precision)
+
+    # Were the model card and license filled?
+    try:
+        license = model_info.cardData["license"]
+    except Exception:
+        return styled_error("Please select a license for your model")
+
+    modelcard_OK, error_msg = check_model_card(model)
+    if not modelcard_OK:
+        return styled_error(error_msg)
+
+    # Seems good, creating the eval
+    print("Adding new eval")
+
+    eval_entry = {
+        "model": model,
+        "base_model": base_model,
+        "revision": revision,
+        "private": private,
+        "precision": precision,
+        "weight_type": weight_type,
+        "status": "PENDING",
+        "submitted_time": current_time,
+        "model_type": model_type,
+        "likes": model_info.likes,
+        "params": model_size,
+        "license": license,
+    }
+
+    # Check for duplicate submission
+    if f"{model}_{revision}_{precision}" in REQUESTED_MODELS:
+        return styled_warning("This model has been already submitted.")
+
+    print("Creating eval file")
+    OUT_DIR = f"{EVAL_REQUESTS_PATH}/{user_name}"
+    os.makedirs(OUT_DIR, exist_ok=True)
+    out_path = f"{OUT_DIR}/{model_path}_eval_request_{private}_{precision}_{weight_type}.json"
+
+    with open(out_path, "w") as f:
+        f.write(json.dumps(eval_entry))
+
+    print("Uploading eval file")
+    API.upload_file(
+        path_or_fileobj=out_path,
+        path_in_repo=out_path.split("eval-queue/")[1],
+        repo_id=QUEUE_REPO,
+        repo_type="dataset",
+        commit_message=f"Add {model} to eval queue",
+    )
+
+    # Remove the local file
+    os.remove(out_path)
+
+    return styled_message(
+        "Your request has been submitted to the evaluation queue!\nPlease wait for up to an hour for the model to show in the PENDING list."
+    )

src/tools/collections.py

@@ -0,0 +1,83 @@
+import os
+
+import pandas as pd
+from huggingface_hub import add_collection_item, delete_collection_item, get_collection, update_collection_item
+from huggingface_hub.utils._errors import HfHubHTTPError
+from pandas import DataFrame
+
+from src.display.utils import AutoEvalColumn, ModelType
+from src.envs import H4_TOKEN, PATH_TO_COLLECTION
+
+# Specific intervals for the collections
+intervals = {
+    "1B": pd.Interval(0, 1.5, closed="right"),
+    "3B": pd.Interval(2.5, 3.5, closed="neither"),
+    "7B": pd.Interval(6, 8, closed="neither"),
+    "13B": pd.Interval(10, 14, closed="neither"),
+    "30B": pd.Interval(25, 35, closed="neither"),
+    "65B": pd.Interval(60, 70, closed="neither"),
+}
+
+
+def update_collections(df: DataFrame):
+    """This function updates the Open LLM Leaderboard model collection with the latest best models for
+    each size category and type.
+    """
+    collection = get_collection(collection_slug=PATH_TO_COLLECTION, token=H4_TOKEN)
+    params_column = pd.to_numeric(df[AutoEvalColumn.params.name], errors="coerce")
+
+    cur_best_models = []
+
+    ix = 0
+    for type in ModelType:
+        if type.value.name == "":
+            continue
+        for size in intervals:
+            # We filter the df to gather the relevant models
+            type_emoji = [t[0] for t in type.value.symbol]
+            filtered_df = df[df[AutoEvalColumn.model_type_symbol.name].isin(type_emoji)]
+
+            numeric_interval = pd.IntervalIndex([intervals[size]])
+            mask = params_column.apply(lambda x: any(numeric_interval.contains(x)))
+            filtered_df = filtered_df.loc[mask]
+
+            best_models = list(
+                filtered_df.sort_values(AutoEvalColumn.average.name, ascending=False)[AutoEvalColumn.dummy.name]
+            )
+            print(type.value.symbol, size, best_models[:10])
+
+            # We add them one by one to the leaderboard
+            for model in best_models:
+                ix += 1
+                cur_len_collection = len(collection.items)
+                try:
+                    collection = add_collection_item(
+                        PATH_TO_COLLECTION,
+                        item_id=model,
+                        item_type="model",
+                        exists_ok=True,
+                        note=f"Best {type.to_str(' ')} model of around {size} on the leaderboard today!",
+                        token=H4_TOKEN,
+                    )
+                    if (
+                        len(collection.items) > cur_len_collection
+                    ):  # we added an item - we make sure its position is correct
+                        item_object_id = collection.items[-1].item_object_id
+                        update_collection_item(
+                            collection_slug=PATH_TO_COLLECTION, item_object_id=item_object_id, position=ix
+                        )
+                        cur_len_collection = len(collection.items)
+                    cur_best_models.append(model)
+                    break
+                except HfHubHTTPError:
+                    continue
+
+    collection = get_collection(PATH_TO_COLLECTION, token=H4_TOKEN)
+    for item in collection.items:
+        if item.item_id not in cur_best_models:
+            try:
+                delete_collection_item(
+                    collection_slug=PATH_TO_COLLECTION, item_object_id=item.item_object_id, token=H4_TOKEN
+                )
+            except HfHubHTTPError:
+                continue

src/tools/model_backlinks.py

@@ -0,0 +1,1309 @@
+models = [
+    "uni-tianyan/Uni-TianYan",
+    "fangloveskari/ORCA_LLaMA_70B_QLoRA",
+    "garage-bAInd/Platypus2-70B-instruct",
+    "upstage/Llama-2-70b-instruct-v2",
+    "fangloveskari/Platypus_QLoRA_LLaMA_70b",
+    "yeontaek/llama-2-70B-ensemble-v5",
+    "TheBloke/Genz-70b-GPTQ",
+    "TheBloke/Platypus2-70B-Instruct-GPTQ",
+    "psmathur/model_007",
+    "yeontaek/llama-2-70B-ensemble-v4",
+    "psmathur/orca_mini_v3_70b",
+    "ehartford/Samantha-1.11-70b",
+    "MayaPH/GodziLLa2-70B",
+    "psmathur/model_007_v2",
+    "chargoddard/MelangeA-70b",
+    "ehartford/Samantha-1.1-70b",
+    "psmathur/model_009",
+    "upstage/Llama-2-70b-instruct",
+    "yeontaek/llama-2-70B-ensemble-v7",
+    "yeontaek/llama-2-70B-ensemble-v6",
+    "chargoddard/MelangeB-70b",
+    "yeontaek/llama-2-70B-ensemble-v3",
+    "chargoddard/MelangeC-70b",
+    "garage-bAInd/Camel-Platypus2-70B",
+    "yeontaek/llama-2-70B-ensemble-v2",
+    "garage-bAInd/Camel-Platypus2-70B",
+    "migtissera/Synthia-70B-v1.2",
+    "v2ray/LLaMA-2-Wizard-70B-QLoRA",
+    "quantumaikr/llama-2-70b-fb16-orca-chat-10k",
+    "v2ray/LLaMA-2-Wizard-70B-QLoRA",
+    "stabilityai/StableBeluga2",
+    "quantumaikr/llama-2-70b-fb16-guanaco-1k",
+    "garage-bAInd/Camel-Platypus2-70B",
+    "migtissera/Synthia-70B-v1.1",
+    "migtissera/Synthia-70B",
+    "psmathur/model_101",
+    "augtoma/qCammel70",
+    "augtoma/qCammel-70",
+    "augtoma/qCammel-70v1",
+    "augtoma/qCammel-70x",
+    "augtoma/qCammel-70-x",
+    "jondurbin/airoboros-l2-70b-gpt4-1.4.1",
+    "dfurman/llama-2-70b-dolphin-peft",
+    "jondurbin/airoboros-l2-70b-2.1",
+    "TheBloke/llama-2-70b-Guanaco-QLoRA-fp16",
+    "quantumaikr/QuantumLM-llama2-70B-Korean-LoRA",
+    "quantumaikr/quantumairk-llama-2-70B-instruct",
+    "psmathur/model_420",
+    "psmathur/model_51",
+    "garage-bAInd/Camel-Platypus2-70B",
+    "TheBloke/Airoboros-L2-70B-2.1-GPTQ",
+    "OpenAssistant/llama2-70b-oasst-sft-v10",
+    "garage-bAInd/Platypus2-70B",
+    "liuxiang886/llama2-70B-qlora-gpt4",
+    "upstage/llama-65b-instruct",
+    "quantumaikr/llama-2-70b-fb16-korean",
+    "NousResearch/Nous-Hermes-Llama2-70b",
+    "v2ray/LLaMA-2-Jannie-70B-QLoRA",
+    "jondurbin/airoboros-l2-70b-gpt4-m2.0",
+    "jondurbin/airoboros-l2-70b-gpt4-m2.0",
+    "OpenAssistant/llama2-70b-oasst-sft-v10",
+    "yeontaek/llama-2-70B-ensemble-v8",
+    "jondurbin/airoboros-l2-70b-gpt4-2.0",
+    "jarradh/llama2_70b_chat_uncensored",
+    "WizardLM/WizardMath-70B-V1.0",
+    "jordiclive/Llama-2-70b-oasst-1-200",
+    "WizardLM/WizardMath-70B-V1.0",
+    "jondurbin/airoboros-l2-70b-gpt4-2.0",
+    "OpenLemur/lemur-70b-chat-v1",
+    "tiiuae/falcon-180B",
+    "tiiuae/falcon-180B",
+    "stabilityai/StableBeluga1-Delta",
+    "psmathur/model_42_70b",
+    "psmathur/test_42_70b",
+    "TheBloke/fiction.live-Kimiko-V2-70B-fp16",
+    "tiiuae/falcon-180B",
+    "WizardLM/WizardMath-70B-V1.0",
+    "tiiuae/falcon-180B-chat",
+    "jondurbin/airoboros-l2-70b-gpt4-2.0",
+    "ehartford/samantha-1.1-llama-33b",
+    "ajibawa-2023/scarlett-33b",
+    "ddobokki/Llama-2-70b-orca-200k",
+    "TheBloke/gpt4-alpaca-lora_mlp-65B-HF",
+    "tiiuae/falcon-180B-chat",
+    "tiiuae/falcon-180B-chat",
+    "tiiuae/falcon-180B",
+    "TheBloke/Lemur-70B-Chat-v1-GPTQ",
+    "NousResearch/Nous-Puffin-70B",
+    "WizardLM/WizardLM-70B-V1.0",
+    "WizardLM/WizardMath-70B-V1.0",
+    "meta-llama/Llama-2-70b-hf",
+    "TheBloke/Llama-2-70B-fp16",
+    "Weyaxi/llama-2-alpacagpt4-1000step",
+    "WizardLM/WizardLM-70B-V1.0",
+    "simsim314/WizardLM-70B-V1.0-HF",
+    "simsim314/WizardLM-70B-V1.0-HF",
+    "WizardLM/WizardLM-70B-V1.0",
+    "openbmb/UltraLM-65b",
+    "psmathur/model_420_preview",
+    "WizardLM/WizardLM-70B-V1.0",
+    "simsim314/WizardLM-70B-V1.0-HF",
+    "OpenBuddy/openbuddy-llama2-70b-v10.1-bf16",
+    "upstage/llama-30b-instruct-2048",
+    "jondurbin/airoboros-65b-gpt4-1.2",
+    "TheBloke/guanaco-65B-HF",
+    "jondurbin/airoboros-65b-gpt4-1.3",
+    "meta-llama/Llama-2-70b-chat-hf",
+    "ValiantLabs/ShiningValiant",
+    "Faradaylab/Aria-70B",
+    "lilloukas/GPlatty-30B",
+    "TheBloke/VicUnlocked-alpaca-65B-QLoRA-fp16",
+    "jondurbin/airoboros-65b-gpt4-1.4-peft",
+    "jondurbin/airoboros-65b-gpt4-1.4",
+    "jondurbin/airoboros-65b-gpt4-2.0",
+    "TheBloke/WizardLM-70B-V1.0-GPTQ",
+    "TheBloke/WizardLM-70B-V1.0-GPTQ",
+    "ariellee/SuperPlatty-30B",
+    "jondurbin/airoboros-65b-gpt4-1.4",
+    "jondurbin/airoboros-65b-gpt4-2.0",
+    "yeontaek/llama-2-70b-IA3-guanaco",
+    "CalderaAI/30B-Lazarus",
+    "Aspik101/trurl-2-13b-pl-instruct_unload",
+    "ehartford/WizardLM-33B-V1.0-Uncensored",
+    "ehartford/WizardLM-33B-V1.0-Uncensored",
+    "OpenBuddy/openbuddy-llama-65b-v8-bf16",
+    "Aspik101/llama-30b-instruct-2048-PL-lora",
+    "h2oai/h2ogpt-research-oasst1-llama-65b",
+    "Aspik101/llama-30b-instruct-2048-PL-lora",
+    "CalderaAI/30B-Epsilon",
+    "Aspik101/llama-30b-2048-instruct-PL-lora_unload",
+    "jondurbin/airoboros-65b-gpt4-m2.0",
+    "jondurbin/airoboros-65b-gpt4-m2.0",
+    "Aeala/Alpaca-elina-65b",
+    "TheBloke/robin-65b-v2-fp16",
+    "TheBloke/gpt4-alpaca-lora-30b-HF",
+    "TheBloke/Llama-2-70B-chat-GPTQ",
+    "upstage/llama-30b-instruct",
+    "OpenLemur/lemur-70b-v1",
+    "lmsys/vicuna-33b-v1.3",
+    "ausboss/llama-30b-supercot",
+    "ai-business/Luban-13B",
+    "Henk717/airochronos-33B",
+    "lmsys/vicuna-33b-v1.3",
+    "Henk717/airochronos-33B",
+    "bavest/fin-llama-33b-merged",
+    "jondurbin/airoboros-33b-gpt4-1.4",
+    "YeungNLP/firefly-llama-30b",
+    "Aspik101/30B-Lazarus-instruct-PL-lora_unload",
+    "uukuguy/speechless-llama2-luban-orca-platypus-13b",
+    "xxyyy123/test_merge_p_ov1_w0.66_w0.5_n1",
+    "jondurbin/airoboros-33b-gpt4-1.2",
+    "TheBloke/alpaca-lora-65B-HF",
+    "bofenghuang/vigogne-33b-instruct",
+    "yeontaek/llama-2-13B-ensemble-v5",
+    "garage-bAInd/Platypus-30B",
+    "Open-Orca/OpenOrca-Platypus2-13B",
+    "kajdun/viwaai-30b_v4",
+    "lilloukas/Platypus-30B",
+    "Open-Orca/OpenOrca-Platypus2-13B",
+    "Henk717/chronoboros-33B",
+    "jondurbin/airoboros-33b-2.1",
+    "HiTZ/alpaca-lora-65b-en-pt-es-ca",
+    "quantumaikr/QuantumLM-70B-hf",
+    "uukuguy/speechless-llama2-13b",
+    "uukuguy/speechless-llama2-hermes-orca-platypus-13b",
+    "openaccess-ai-collective/manticore-30b-chat-pyg-alpha",
+    "LLMs/WizardLM-30B-V1.0",
+    "TheBloke/WizardLM-30B-fp16",
+    "openaccess-ai-collective/hippogriff-30b-chat",
+    "concedo/Vicuzard-30B-Uncensored",
+    "TFLai/OpenOrca-Platypus2-13B-QLoRA-0.80-epoch",
+    "huggingface/llama-65b",
+    "huggyllama/llama-65b",
+    "gaodrew/gaodrew-llama-30b-instruct-2048-Open-Platypus-100steps",
+    "uukuguy/speechless-llama2-hermes-orca-platypus-wizardlm-13b",
+    "Sao10K/Mythical-Destroyer-V2-L2-13B",
+    "camel-ai/CAMEL-33B-Combined-Data",
+    "dsvv-cair/alpaca-cleaned-llama-30b-bf16",
+    "MetaIX/GPT4-X-Alpasta-30b",
+    "garage-bAInd/Stable-Platypus2-13B",
+    "TFLai/Luban-Platypus2-13B-QLora-0.80-epoch",
+    "TheBloke/OpenOrca-Platypus2-13B-GPTQ",
+    "IkariDev/Athena-tmp",
+    "OpenBuddyEA/openbuddy-llama-30b-v7.1-bf16",
+    "OpenBuddyEA/openbuddy-llama-30b-v7.1-bf16",
+    "Open-Orca/OpenOrcaxOpenChat-Preview2-13B",
+    "psmathur/model_007_13b_v2",
+    "Aspik101/Vicuzard-30B-Uncensored-instruct-PL-lora_unload",
+    "jondurbin/airoboros-33b-gpt4-m2.0",
+    "Sao10K/Mythical-Destroyer-L2-13B",
+    "TheBloke/Wizard-Vicuna-30B-Uncensored-fp16",
+    "ehartford/Wizard-Vicuna-30B-Uncensored",
+    "TFLai/Nova-13B",
+    "TheBloke/robin-33B-v2-fp16",
+    "totally-not-an-llm/PuddleJumper-13b",
+    "Aeala/VicUnlocked-alpaca-30b",
+    "Yhyu13/oasst-rlhf-2-llama-30b-7k-steps-hf",
+    "jondurbin/airoboros-33b-gpt4",
+    "jondurbin/airoboros-33b-gpt4-m2.0",
+    "tiiuae/falcon-40b-instruct",
+    "psmathur/orca_mini_v3_13b",
+    "Aeala/GPT4-x-AlpacaDente-30b",
+    "MayaPH/GodziLLa-30B",
+    "jondurbin/airoboros-33b-gpt4-m2.0",
+    "TFLai/SpeechlessV1-Nova-13B",
+    "yeontaek/llama-2-13B-ensemble-v4",
+    "ajibawa-2023/carl-33b",
+    "jondurbin/airoboros-33b-gpt4-2.0",
+    "TFLai/Stable-Platypus2-13B-QLoRA-0.80-epoch",
+    "jondurbin/airoboros-33b-gpt4-1.3",
+    "TehVenom/oasst-sft-6-llama-33b-xor-MERGED-16bit",
+    "TFLai/OrcaMini-Platypus2-13B-QLoRA-0.80-epoch",
+    "jondurbin/airoboros-33b-gpt4-2.0",
+    "chargoddard/Chronorctypus-Limarobormes-13b",
+    "jondurbin/airoboros-33b-gpt4-1.3",
+    "Open-Orca/OpenOrca-Platypus2-13B",
+    "FelixChao/vicuna-33b-coder",
+    "FelixChao/vicuna-33b-coder",
+    "Gryphe/MythoMix-L2-13b",
+    "Aeala/Enterredaas-33b",
+    "yeontaek/llama-2-13B-ensemble-v1",
+    "TFLai/OpenOrcaPlatypus2-Platypus2-13B-QLora-0.80-epoch",
+    "TFLai/Ensemble5-Platypus2-13B-QLora-0.80-epoch",
+    "yeontaek/llama-2-13B-ensemble-v3",
+    "TFLai/MythoMix-Platypus2-13B-QLoRA-0.80-epoch",
+    "yihan6324/llama2-13b-instructmining-40k-sharegpt",
+    "timdettmers/guanaco-33b-merged",
+    "TFLai/EnsembleV5-Nova-13B",
+    "circulus/Llama-2-13b-orca-v1",
+    "Undi95/ReMM-SLERP-L2-13B",
+    "Gryphe/MythoMax-L2-13b",
+    "stabilityai/StableBeluga-13B",
+    "circulus/Llama-2-13b-orca-v1",
+    "ehartford/WizardLM-30B-Uncensored",
+    "The-Face-Of-Goonery/huginnv1.2",
+    "TheBloke/OpenOrcaxOpenChat-Preview2-13B-GPTQ",
+    "Sao10K/Stheno-L2-13B",
+    "bofenghuang/vigogne-2-13b-instruct",
+    "The-Face-Of-Goonery/Huginn-13b-FP16",
+    "grimpep/L2-MythoMax22b-instruct-Falseblock",
+    "TFLai/Nous-Hermes-Platypus2-13B-QLoRA-0.80-epoch",
+    "yeontaek/Platypus2xOpenOrca-13B-IA3-v4",
+    "yeontaek/Platypus2xOpenOrca-13B-IA3",
+    "yeontaek/Platypus2xOpenOrca-13B-IA3-ensemble",
+    "Open-Orca/LlongOrca-13B-16k",
+    "Sao10K/Stheno-Inverted-L2-13B",
+    "garage-bAInd/Camel-Platypus2-13B",
+    "digitous/Alpacino30b",
+    "NousResearch/Nous-Hermes-Llama2-13b",
+    "yeontaek/Platypus2xOpenOrca-13B-IA3-v3",
+    "TFLai/MythicalDestroyerV2-Platypus2-13B-QLora-0.80-epoch",
+    "TheBloke/VicUnlocked-30B-LoRA-HF",
+    "Undi95/Nous-Hermes-13B-Code",
+    "The-Face-Of-Goonery/Chronos-Beluga-v2-13bfp16",
+    "NousResearch/Nous-Hermes-Llama2-13b",
+    "Monero/WizardLM-Uncensored-SuperCOT-StoryTelling-30b",
+    "TheBloke/Wizard-Vicuna-30B-Uncensored-GPTQ",
+    "Open-Orca/OpenOrcaxOpenChat-Preview2-13B",
+    "Austism/chronos-hermes-13b-v2",
+    "yeontaek/Platypus2xOpenOrca-13B-IA3-v2.1",
+    "yeontaek/Platypus2xOpenOrca-13B-IA3-v2",
+    "Gryphe/MythoLogic-L2-13b",
+    "augtoma/qCammel-13",
+    "YeungNLP/firefly-llama2-13b-v1.2",
+    "Aspik101/StableBeluga-13B-instruct-PL-lora_unload",
+    "andreaskoepf/llama2-13b-megacode2_min100",
+    "rombodawg/LosslessMegaCoder-llama2-13b-mini",
+    "yulan-team/YuLan-Chat-2-13b-fp16",
+    "elinas/chronos-33b",
+    "YeungNLP/firefly-llama2-13b",
+    "Sao10K/Medusa-13b",
+    "OptimalScale/robin-65b-v2-delta",
+    "minlik/chinese-alpaca-33b-merged",
+    "OpenAssistant/llama2-13b-megacode2-oasst",
+    "TheBloke/OpenAssistant-SFT-7-Llama-30B-HF",
+    "Undi95/UndiMix-v1-13b",
+    "ehartford/Samantha-1.11-13b",
+    "beaugogh/Llama2-13b-sharegpt4",
+    "Aeala/GPT4-x-AlpacaDente2-30b",
+    "luffycodes/nash-vicuna-13b-v1dot5-ep2-w-rag-w-simple",
+    "WizardLM/WizardLM-13B-V1.1",
+    "uukuguy/speechless-orca-platypus-coig-lite-2k-0.6e-13b",
+    "huggyllama/llama-30b",
+    "Undi95/ReMM-L2-13B-PIPPA",
+    "Undi95/ReMM-L2-13B",
+    "gaodrew/gaodrew-gorgonzola-13b",
+    "lmsys/vicuna-13b-v1.5",
+    "yeontaek/Platypus2xOpenOrca-13B-LoRa",
+    "Yhyu13/llama-30B-hf-openassitant",
+    "huggingface/llama-30b",
+    "lmsys/vicuna-13b-v1.5",
+    "TFLai/Athena-Platypus2-13B-QLora-0.80-epoch",
+    "TheBloke/dromedary-65b-lora-HF",
+    "yeontaek/llama-2-13b-Beluga-QLoRA",
+    "The-Face-Of-Goonery/Huginn-13b-V4",
+    "The-Face-Of-Goonery/Huginn-13b-v4.5",
+    "The-Face-Of-Goonery/Huginn-v3-13b",
+    "tiiuae/falcon-40b",
+    "WhoTookMyAmogusNickname/NewHope_HF_not_official",
+    "gaodrew/OpenOrca-Platypus2-13B-thera-1250",
+    "SLAM-group/NewHope",
+    "garage-bAInd/Platypus2-13B",
+    "migtissera/Synthia-13B",
+    "elinas/chronos-13b-v2",
+    "mosaicml/mpt-30b-chat",
+    "CHIH-HUNG/llama-2-13b-OpenOrca_5w",
+    "uukuguy/speechless-hermes-coig-lite-13b",
+    "TheBloke/tulu-30B-fp16",
+    "uukuguy/speechless-hermes-coig-lite-13b",
+    "xDAN-AI/xDAN_13b_l2_lora",
+    "lmsys/vicuna-13b-v1.5-16k",
+    "openchat/openchat_v3.1",
+    "CHIH-HUNG/llama-2-13b-dolphin_5w",
+    "Aspik101/vicuna-13b-v1.5-PL-lora_unload",
+    "Undi95/MLewd-L2-13B",
+    "ehartford/minotaur-llama2-13b-qlora",
+    "kajdun/iubaris-13b-v3",
+    "TFLai/Limarp-Platypus2-13B-QLoRA-0.80-epoch",
+    "openchat/openchat_v3.1",
+    "uukuguy/speechless-orca-platypus-coig-lite-4k-0.6e-13b",
+    "ziqingyang/chinese-alpaca-2-13b",
+    "TFLai/Airboros2.1-Platypus2-13B-QLora-0.80-epoch",
+    "yeontaek/llama-2-13b-Guanaco-QLoRA",
+    "lmsys/vicuna-13b-v1.5-16k",
+    "ehartford/based-30b",
+    "kingbri/airolima-chronos-grad-l2-13B",
+    "openchat/openchat_v3.2",
+    "uukuguy/speechless-orca-platypus-coig-lite-4k-0.5e-13b",
+    "yeontaek/Platypus2-13B-LoRa",
+    "kingbri/chronolima-airo-grad-l2-13B",
+    "openchat/openchat_v3.2",
+    "TFLai/PuddleJumper-Platypus2-13B-QLoRA-0.80-epoch",
+    "shareAI/llama2-13b-Chinese-chat",
+    "ehartford/WizardLM-1.0-Uncensored-Llama2-13b",
+    "Aspik101/Redmond-Puffin-13B-instruct-PL-lora_unload",
+    "yeontaek/llama-2-13B-ensemble-v6",
+    "WizardLM/WizardLM-13B-V1.2",
+    "TheBloke/WizardLM-13B-V1.1-GPTQ",
+    "bhenrym14/airophin-13b-pntk-16k-fp16",
+    "ehartford/WizardLM-1.0-Uncensored-Llama2-13b",
+    "Mikael110/llama-2-13b-guanaco-fp16",
+    "yeontaek/airoboros-2.1-llama-2-13B-QLoRa",
+    "CalderaAI/13B-Legerdemain-L2",
+    "grimpep/llama2-22b-wizard_vicuna",
+    "grimpep/llama2-22B-GPLATTY",
+    "bhenrym14/airophin-13b-pntk-16k-fp16",
+    "yeontaek/llama-2-13b-QLoRA",
+    "OpenAssistant/llama2-13b-orca-8k-3319",
+    "TheBloke/WizardLM-13B-V1-1-SuperHOT-8K-fp16",
+    "duliadotio/dulia-13b-8k-alpha",
+    "Undi95/LewdEngine",
+    "OpenBuddy/openbuddy-llama2-13b-v8.1-fp16",
+    "CHIH-HUNG/llama-2-13b-open_orca_20w",
+    "bhenrym14/airoboros-33b-gpt4-1.4.1-lxctx-PI-16384-fp16",
+    "FlagAlpha/Llama2-Chinese-13b-Chat",
+    "LLMs/WizardLM-13B-V1.0",
+    "chansung/gpt4-alpaca-lora-13b-decapoda-1024",
+    "TheBloke/wizardLM-13B-1.0-fp16",
+    "digitous/13B-Chimera",
+    "yeontaek/Platypus2xOpenOrcaxGuanaco-13B-LoRa",
+    "jondurbin/airoboros-l2-13b-2.1",
+    "Monero/WizardLM-30B-Uncensored-Guanaco-SuperCOT-30b",
+    "TheBloke/UltraLM-13B-fp16",
+    "openaccess-ai-collective/minotaur-13b-fixed",
+    "NousResearch/Redmond-Puffin-13B",
+    "KoboldAI/LLaMA2-13B-Holomax",
+    "Lajonbot/WizardLM-13B-V1.2-PL-lora_unload",
+    "yeontaek/Platypus2-13B-LoRa-v2",
+    "TheBloke/airoboros-13B-HF",
+    "jondurbin/airoboros-13b",
+    "jjaaaww/posi_13b",
+    "CoolWP/llama-2-13b-guanaco-fp16",
+    "yeontaek/Platypus2-13B-QLoRa",
+    "h2oai/h2ogpt-research-oig-oasst1-512-30b",
+    "dfurman/llama-2-13b-guanaco-peft",
+    "NousResearch/Redmond-Puffin-13B",
+    "pe-nlp/llama-2-13b-platypus-vicuna-wizard",
+    "CHIH-HUNG/llama-2-13b-dolphin_20w",
+    "NousResearch/Nous-Hermes-13b",
+    "NobodyExistsOnTheInternet/GiftedConvo13bLoraNoEconsE4",
+    "ehartford/Wizard-Vicuna-13B-Uncensored",
+    "TheBloke/Wizard-Vicuna-13B-Uncensored-HF",
+    "openchat/openchat_v3.2_super",
+    "bhenrym14/airophin-v2-13b-PI-8k-fp16",
+    "openaccess-ai-collective/manticore-13b",
+    "The-Face-Of-Goonery/Huginn-22b-Prototype",
+    "jphme/Llama-2-13b-chat-german",
+    "grimpep/llama2-28B-Airo03",
+    "TheBloke/Kimiko-v2-13B-fp16",
+    "FPHam/Free_Sydney_13b_HF",
+    "lmsys/vicuna-13b-v1.3",
+    "FelixChao/llama2-13b-math1.1",
+    "CalderaAI/13B-BlueMethod",
+    "meta-llama/Llama-2-13b-chat-hf",
+    "deepse/CodeUp-Llama-2-13b-chat-hf",
+    "WizardLM/WizardMath-13B-V1.0",
+    "WizardLM/WizardMath-13B-V1.0",
+    "HyperbeeAI/Tulpar-7b-v0",
+    "xxyyy123/test_qkvo_adptor",
+    "xxyyy123/mc_data_30k_from_platpus_orca_7b_10k_v1_lora_qkvo_rank14_v2",
+    "openchat/openchat_v2_w",
+    "FelixChao/llama2-13b-math1.1",
+    "psmathur/orca_mini_v3_7b",
+    "TehVenom/Metharme-13b-Merged",
+    "xxyyy123/10k_v1_lora_qkvo_rank14_v3",
+    "OpenAssistant/llama2-13b-orca-v2-8k-3166",
+    "openaccess-ai-collective/wizard-mega-13b",
+    "jondurbin/airoboros-13b-gpt4-1.4",
+    "jondurbin/airoboros-13b-gpt4-1.4-fp16",
+    "Monero/Manticore-13b-Chat-Pyg-Guanaco",
+    "FelixChao/llama2-13b-math1.2",
+    "chargoddard/platypus-2-22b-relora",
+    "FelixChao/llama2-13b-math1.2",
+    "Gryphe/MythoBoros-13b",
+    "CalderaAI/13B-Ouroboros",
+    "OpenAssistant/llama2-13b-orca-v2-8k-3166",
+    "heegyu/LIMA2-13b-hf",
+    "digitous/13B-HyperMantis",
+    "Gryphe/MythoLogic-13b",
+    "TheBloke/Airoboros-L2-13B-2.1-GPTQ",
+    "chargoddard/platypus2-22b-relora",
+    "openchat/openchat_v2",
+    "yeontaek/Platypus2-13B-IA3",
+    "stabilityai/StableBeluga-7B",
+    "circulus/Llama-2-7b-orca-v1",
+    "budecosystem/genz-13b-v2",
+    "TheBloke/gpt4-x-vicuna-13B-HF",
+    "NobodyExistsOnTheInternet/GiftedConvo13bLoraNoEcons",
+    "zarakiquemparte/zarafusionex-1.1-l2-7b",
+    "Lajonbot/tableBeluga-7B-instruct-pl-lora_unload",
+    "jondurbin/airoboros-13b-gpt4",
+    "gaodrew/gaodrew-gorgonzola-13b",
+    "jondurbin/airoboros-13b-gpt4-1.1",
+    "TheBloke/gpt4-alpaca-lora-13B-HF",
+    "zarakiquemparte/zarablendex-vq-l2-7b",
+    "openaccess-ai-collective/manticore-13b-chat-pyg",
+    "Lajonbot/Llama-2-13b-hf-instruct-pl-lora_unload",
+    "NobodyExistsOnTheInternet/PuffedLIMA13bQLORA",
+    "xxyyy123/10k_v1_lora_qkvo_rank28_v2",
+    "jondurbin/airoboros-l2-13b-gpt4-1.4.1",
+    "dhmeltzer/Llama-2-13b-hf-eli5-wiki-1024_r_64_alpha_16",
+    "NobodyExistsOnTheInternet/PuffedConvo13bLoraE4",
+    "yihan6324/llama2-7b-instructmining-40k-sharegpt",
+    "CHIH-HUNG/llama-2-13b-Open_Platypus_and_ccp_2.6w",
+    "Aeala/GPT4-x-Alpasta-13b",
+    "psmathur/orca_mini_v2_13b",
+    "YeungNLP/firefly-llama-13b",
+    "psmathur/orca_mini_v2_13b",
+    "zarakiquemparte/zarafusionix-l2-7b",
+    "yihan6324/llama2-7b-instructmining-60k-sharegpt",
+    "yihan6324/llama-2-7b-instructmining-60k-sharegpt",
+    "layoric/llama-2-13b-code-alpaca",
+    "bofenghuang/vigogne-13b-instruct",
+    "Lajonbot/vicuna-13b-v1.3-PL-lora_unload",
+    "lvkaokao/llama2-7b-hf-chat-lora-v3",
+    "ehartford/dolphin-llama-13b",
+    "YeungNLP/firefly-llama-13b-v1.2",
+    "TheBloke/Kimiko-13B-fp16",
+    "kevinpro/Vicuna-13B-CoT",
+    "eachadea/vicuna-13b-1.1",
+    "pillowtalks-ai/delta13b",
+    "TheBloke/vicuna-13B-1.1-HF",
+    "TheBloke/Vicuna-13B-CoT-fp16",
+    "lmsys/vicuna-13b-delta-v1.1",
+    "lmsys/vicuna-13b-v1.1",
+    "xxyyy123/20k_v1_lora_qkvo_rank14_v2",
+    "TheBloke/guanaco-13B-HF",
+    "TheBloke/vicuna-13b-v1.3.0-GPTQ",
+    "edor/Stable-Platypus2-mini-7B",
+    "totally-not-an-llm/EverythingLM-13b-V2-16k",
+    "zarakiquemparte/zaraxe-l2-7b",
+    "beaugogh/Llama2-7b-openorca-mc-v2",
+    "TheBloke/Nous-Hermes-13B-SuperHOT-8K-fp16",
+    "quantumaikr/QuantumLM",
+    "jondurbin/airoboros-13b-gpt4-1.2",
+    "TheBloke/robin-13B-v2-fp16",
+    "TFLai/llama-2-13b-4bit-alpaca-gpt4",
+    "yihan6324/llama2-7b-instructmining-orca-40k",
+    "dvruette/oasst-llama-13b-2-epochs",
+    "Open-Orca/LlongOrca-7B-16k",
+    "Aspik101/Nous-Hermes-13b-pl-lora_unload",
+    "ehartford/Samantha-1.11-CodeLlama-34b",
+    "nkpz/llama2-22b-chat-wizard-uncensored",
+    "bofenghuang/vigogne-13b-chat",
+    "beaugogh/Llama2-7b-openorca-mc-v1",
+    "OptimalScale/robin-13b-v2-delta",
+    "pe-nlp/llama-2-13b-vicuna-wizard",
+    "chargoddard/llama2-22b",
+    "gywy/llama2-13b-chinese-v1",
+    "frank098/Wizard-Vicuna-13B-juniper",
+    "IGeniusDev/llama13B-quant8-testv1-openorca-customdataset",
+    "CHIH-HUNG/llama-2-13b-huangyt_Fintune_1_17w-gate_up_down_proj",
+    "eachadea/vicuna-13b",
+    "yihan6324/llama2-7b-instructmining-orca-90k",
+    "chargoddard/llama2-22b-blocktriangular",
+    "luffycodes/mcq-vicuna-13b-v1.5",
+    "Yhyu13/chimera-inst-chat-13b-hf",
+    "luffycodes/mcq-vicuna-13b-v1.5",
+    "chargoddard/ypotryll-22b-epoch2-qlora",
+    "totally-not-an-llm/EverythingLM-13b-16k",
+    "luffycodes/mcq-hal-vicuna-13b-v1.5",
+    "openaccess-ai-collective/minotaur-13b",
+    "IGeniusDev/llama13B-quant8-testv1-openorca-customdataset",
+    "chargoddard/llama2-22b-blocktriangular",
+    "TFLai/Platypus2-13B-QLoRA-0.80-epoch",
+    "meta-llama/Llama-2-13b-hf",
+    "CHIH-HUNG/llama-2-13b-huangyt_FINETUNE2_3w-gate_up_down_proj",
+    "luffycodes/mcq-hal-vicuna-13b-v1.5",
+    "TheBloke/Llama-2-13B-fp16",
+    "TaylorAI/Flash-Llama-13B",
+    "shareAI/bimoGPT-llama2-13b",
+    "wahaha1987/llama_13b_sharegpt94k_fastchat",
+    "openchat/openchat_8192",
+    "CHIH-HUNG/llama-2-13b-huangyt_Fintune_1_17w-q_k_v_o_proj",
+    "dvruette/llama-13b-pretrained-sft-do2",
+    "CHIH-HUNG/llama-2-13b-alpaca-test",
+    "OpenBuddy/openbuddy-llama2-13b-v11.1-bf16",
+    "CHIH-HUNG/llama-2-13b-FINETUNE2_TEST_2.2w",
+    "project-baize/baize-v2-13b",
+    "jondurbin/airoboros-l2-13b-gpt4-m2.0",
+    "yeontaek/Platypus2xOpenOrca-13B-LoRa-v2",
+    "CHIH-HUNG/llama-2-13b-huangyt_FINETUNE2_3w",
+    "xzuyn/Alpacino-SuperCOT-13B",
+    "jondurbin/airoboros-l2-13b-gpt4-2.0",
+    "aiplanet/effi-13b",
+    "clibrain/Llama-2-13b-ft-instruct-es",
+    "CHIH-HUNG/llama-2-13b-huangyt_Fintune_1_17w",
+    "bofenghuang/vigogne-2-7b-instruct",
+    "CHIH-HUNG/llama-2-13b-huangyt_FINETUNE2_3w-q_k_v_o_proj",
+    "bofenghuang/vigogne-2-7b-chat",
+    "aiplanet/effi-13b",
+    "haonan-li/bactrian-x-llama-13b-merged",
+    "beaugogh/Llama2-7b-sharegpt4",
+    "HWERI/Llama2-7b-sharegpt4",
+    "jondurbin/airoboros-13b-gpt4-1.3",
+    "jondurbin/airoboros-c34b-2.1",
+    "junelee/wizard-vicuna-13b",
+    "TheBloke/wizard-vicuna-13B-HF",
+    "Open-Orca/OpenOrca-Preview1-13B",
+    "TheBloke/h2ogpt-oasst1-512-30B-HF",
+    "TheBloke/Llama-2-13B-GPTQ",
+    "camel-ai/CAMEL-13B-Combined-Data",
+    "lmsys/vicuna-7b-v1.5",
+    "lmsys/vicuna-7b-v1.5-16k",
+    "lmsys/vicuna-7b-v1.5",
+    "ausboss/llama-13b-supercot",
+    "TheBloke/tulu-13B-fp16",
+    "NousResearch/Nous-Hermes-llama-2-7b",
+    "jlevin/guanaco-13b-llama-2",
+    "lmsys/vicuna-7b-v1.5-16k",
+    "dvruette/llama-13b-pretrained",
+    "nkpz/llama2-22b-daydreamer-v3",
+    "dvruette/llama-13b-pretrained-dropout",
+    "jondurbin/airoboros-l2-13b-2.1",
+    "LLMs/Stable-Vicuna-13B",
+    "64bits/LexPodLM-13B",
+    "lizhuang144/llama_mirror_13b_v1.0",
+    "TheBloke/stable-vicuna-13B-HF",
+    "zarakiquemparte/zaraxls-l2-7b",
+    "TheBloke/Llama-2-13B-GPTQ",
+    "Kiddyz/testlm-3",
+    "migtissera/Synthia-7B",
+    "zarakiquemparte/zarablend-l2-7b",
+    "mosaicml/mpt-30b-instruct",
+    "PocketDoc/Dans-PileOfSets-Mk1-llama-13b-merged",
+    "vonjack/Qwen-LLaMAfied-HFTok-7B-Chat",
+    "l3utterfly/llama2-7b-layla",
+    "Lajonbot/vicuna-7b-v1.5-PL-lora_unload",
+    "heegyu/LIMA-13b-hf",
+    "frank098/WizardLM_13B_juniper",
+    "ashercn97/manatee-7b",
+    "chavinlo/gpt4-x-alpaca",
+    "PocketDoc/Dans-PersonalityEngine-13b",
+    "ehartford/WizardLM-1.0-Uncensored-CodeLlama-34b",
+    "digitous/Alpacino13b",
+    "edor/Hermes-Platypus2-mini-7B",
+    "lvkaokao/llama2-7b-hf-chat-lora-v2",
+    "Kiddyz/testlm-1-1",
+    "Kiddyz/testlm",
+    "Kiddyz/testlm-1",
+    "Kiddyz/testlm2",
+    "radm/Philosophy-Platypus2-13b",
+    "aiplanet/effi-13b",
+    "Harshvir/Llama-2-7B-physics",
+    "YeungNLP/firefly-ziya-13b",
+    "LinkSoul/Chinese-Llama-2-7b",
+    "PeanutJar/LLaMa-2-PeanutButter_v10-7B",
+    "OpenBuddy/openbuddy-llama2-13b-v11-bf16",
+    "StudentLLM/Alpagasus-2-13B-QLoRA-pipeline",
+    "meta-llama/Llama-2-13b-hf",
+    "WizardLM/WizardCoder-Python-34B-V1.0",
+    "dvruette/llama-13b-pretrained-sft-epoch-1",
+    "camel-ai/CAMEL-13B-Role-Playing-Data",
+    "ziqingyang/chinese-llama-2-13b",
+    "rombodawg/LosslessMegaCoder-llama2-7b-mini",
+    "TheBloke/koala-13B-HF",
+    "lmsys/vicuna-7b-delta-v1.1",
+    "eachadea/vicuna-7b-1.1",
+    "Ejafa/vicuna_7B_vanilla_1.1",
+    "lvkaokao/llama2-7b-hf-chat-lora",
+    "OpenBuddy/openbuddy-atom-13b-v9-bf16",
+    "Norquinal/llama-2-7b-claude-chat-rp",
+    "Danielbrdz/Barcenas-7b",
+    "heegyu/WizardVicuna2-13b-hf",
+    "meta-llama/Llama-2-7b-chat-hf",
+    "PeanutJar/LLaMa-2-PeanutButter_v14-7B",
+    "PeanutJar/LLaMa-2-PeanutButter_v4-7B",
+    "davzoku/cria-llama2-7b-v1.3",
+    "OpenBuddy/openbuddy-atom-13b-v9-bf16",
+    "lvkaokao/llama2-7b-hf-instruction-lora",
+    "Tap-M/Luna-AI-Llama2-Uncensored",
+    "ehartford/Samantha-1.11-7b",
+    "WizardLM/WizardCoder-Python-34B-V1.0",
+    "TheBloke/Manticore-13B-Chat-Pyg-Guanaco-SuperHOT-8K-GPTQ",
+    "Mikael110/llama-2-7b-guanaco-fp16",
+    "garage-bAInd/Platypus2-7B",
+    "PeanutJar/LLaMa-2-PeanutButter_v18_B-7B",
+    "mosaicml/mpt-30b",
+    "garage-bAInd/Platypus2-7B",
+    "huggingface/llama-13b",
+    "dvruette/oasst-llama-13b-1000-steps",
+    "jordiclive/gpt4all-alpaca-oa-codealpaca-lora-13b",
+    "huggyllama/llama-13b",
+    "Voicelab/trurl-2-7b",
+    "TFLai/llama-13b-4bit-alpaca",
+    "gywy/llama2-13b-chinese-v2",
+    "lmsys/longchat-13b-16k",
+    "Aspik101/trurl-2-7b-pl-instruct_unload",
+    "WizardLM/WizardMath-7B-V1.0",
+    "Norquinal/llama-2-7b-claude-chat",
+    "TheTravellingEngineer/llama2-7b-chat-hf-dpo",
+    "HuggingFaceH4/starchat-beta",
+    "joehuangx/spatial-vicuna-7b-v1.5-LoRA",
+    "conceptofmind/LLongMA-2-13b-16k",
+    "tianyil1/denas-llama2",
+    "lmsys/vicuna-7b-v1.3",
+    "conceptofmind/LLongMA-2-13b-16k",
+    "openchat/opencoderplus",
+    "ajibawa-2023/scarlett-7b",
+    "dhmeltzer/llama-7b-SFT_eli5_wiki65k_1024_r_64_alpha_16_merged",
+    "psyche/kollama2-7b-v2",
+    "heegyu/LIMA2-7b-hf",
+    "dhmeltzer/llama-7b-SFT-qlora-eli5-wiki_DPO_ds_RM_top_2_1024_r_64_alpha_16",
+    "abhishek/llama2guanacotest",
+    "jondurbin/airoboros-l2-7b-2.1",
+    "llama-anon/instruct-13b",
+    "FelixChao/vicuna-7B-physics",
+    "Aspik101/Llama-2-7b-hf-instruct-pl-lora_unload",
+    "shibing624/chinese-alpaca-plus-13b-hf",
+    "davzoku/cria-llama2-7b-v1.3_peft",
+    "quantumaikr/llama-2-7b-hf-guanaco-1k",
+    "togethercomputer/Llama-2-7B-32K-Instruct",
+    "sia-ai/llama-2-7b-1-percent-open-orca-1000-steps-v0",
+    "TheTravellingEngineer/llama2-7b-hf-guanaco",
+    "Lajonbot/Llama-2-7b-chat-hf-instruct-pl-lora_unload",
+    "jondurbin/airoboros-l2-7b-gpt4-1.4.1",
+    "wahaha1987/llama_7b_sharegpt94k_fastchat",
+    "FelixChao/vicuna-7B-chemical",
+    "TinyPixel/llama2-7b-oa",
+    "chaoyi-wu/MedLLaMA_13B",
+    "edor/Platypus2-mini-7B",
+    "RoversX/llama-2-7b-hf-small-shards-Samantha-V1-SFT",
+    "venkycs/llama-v2-7b-32kC-Security",
+    "psyche/kollama2-7b",
+    "Fredithefish/Guanaco-7B-Uncensored",
+    "TheTravellingEngineer/llama2-7b-chat-hf-guanaco",
+    "ehartford/WizardLM-13B-Uncensored",
+    "PocketDoc/Dans-CreepingSenseOfDoom",
+    "wenge-research/yayi-7b-llama2",
+    "georgesung/llama2_7b_chat_uncensored",
+    "TinyPixel/llama2-7b-instruct",
+    "quantumaikr/QuantumLM-7B",
+    "xzuyn/MedicWizard-7B",
+    "wenge-research/yayi-7b-llama2",
+    "TinyPixel/lima-test",
+    "elyza/ELYZA-japanese-Llama-2-7b-instruct",
+    "lgaalves/llama-2-7b-hf_open-platypus",
+    "ziqingyang/chinese-alpaca-2-7b",
+    "TehVenom/Pygmalion-Vicuna-1.1-7b",
+    "meta-llama/Llama-2-7b-hf",
+    "bongchoi/test-llama2-7b",
+    "TaylorAI/Flash-Llama-7B",
+    "TheTravellingEngineer/llama2-7b-chat-hf-v2",
+    "TheTravellingEngineer/llama2-7b-chat-hf-v4",
+    "kashif/stack-llama-2",
+    "PeanutJar/LLaMa-2-PeanutButter_v18_A-7B",
+    "ToolBench/ToolLLaMA-7b-LoRA",
+    "Monero/WizardLM-13b-OpenAssistant-Uncensored",
+    "TheTravellingEngineer/llama2-7b-chat-hf-v2",
+    "TheTravellingEngineer/llama2-7b-chat-hf-v4",
+    "mrm8488/llama-2-coder-7b",
+    "elyza/ELYZA-japanese-Llama-2-7b-fast-instruct",
+    "clibrain/Llama-2-7b-ft-instruct-es",
+    "medalpaca/medalpaca-7b",
+    "TheBloke/tulu-7B-fp16",
+    "OpenBuddy/openbuddy-openllama-13b-v7-fp16",
+    "TaylorAI/FLAN-Llama-7B-2_Llama2-7B-Flash_868_full_model",
+    "Aspik101/vicuna-7b-v1.3-instruct-pl-lora_unload",
+    "jondurbin/airoboros-l2-7b-gpt4-2.0",
+    "dhmeltzer/llama-7b-SFT_ds_eli5_1024_r_64_alpha_16_merged",
+    "GOAT-AI/GOAT-7B-Community",
+    "AtomEchoAI/AtomGPT_56k",
+    "julianweng/Llama-2-7b-chat-orcah",
+    "TehVenom/Pygmalion-13b-Merged",
+    "jondurbin/airoboros-7b-gpt4-1.1",
+    "dhmeltzer/llama-7b-SFT_ds_wiki65k_1024_r_64_alpha_16_merged",
+    "bofenghuang/vigogne-7b-chat",
+    "lmsys/longchat-7b-v1.5-32k",
+    "jondurbin/airoboros-l2-7b-gpt4-m2.0",
+    "synapsoft/Llama-2-7b-chat-hf-flan2022-1.2M",
+    "jondurbin/airoboros-7b-gpt4-1.4",
+    "Charlie911/vicuna-7b-v1.5-lora-mctaco",
+    "yihan6324/instructmining-platypus-15k",
+    "meta-llama/Llama-2-7b-hf",
+    "TheTravellingEngineer/llama2-7b-chat-hf-v3",
+    "quantumaikr/KoreanLM-hf",
+    "openthaigpt/openthaigpt-1.0.0-alpha-7b-chat-ckpt-hf",
+    "TheBloke/Llama-2-7B-GPTQ",
+    "TheBloke/Llama-2-7B-GPTQ",
+    "LLMs/AlpacaGPT4-7B-elina",
+    "ehartford/Wizard-Vicuna-7B-Uncensored",
+    "TheBloke/Wizard-Vicuna-7B-Uncensored-HF",
+    "TheTravellingEngineer/llama2-7b-chat-hf-v3",
+    "golaxy/gowizardlm",
+    "ehartford/dolphin-llama2-7b",
+    "CHIH-HUNG/llama-2-7b-dolphin_10w-test",
+    "mncai/chatdoctor",
+    "psyche/kollama2-7b-v3",
+    "jondurbin/airoboros-7b-gpt4",
+    "jondurbin/airoboros-7b",
+    "TheBloke/airoboros-7b-gpt4-fp16",
+    "mosaicml/mpt-7b-8k-chat",
+    "elyza/ELYZA-japanese-Llama-2-7b",
+    "bofenghuang/vigogne-7b-instruct",
+    "jxhong/CAlign-alpaca-7b",
+    "golaxy/goims",
+    "jondurbin/airoboros-7b-gpt4-1.2",
+    "jphme/orca_mini_v2_ger_7b",
+    "psmathur/orca_mini_v2_7b",
+    "notstoic/PygmalionCoT-7b",
+    "golaxy/gogpt2-13b",
+    "golaxy/gogpt2-13b-chat",
+    "togethercomputer/LLaMA-2-7B-32K",
+    "TheBloke/wizardLM-7B-HF",
+    "keyfan/vicuna-chinese-replication-v1.1",
+    "golaxy/gogpt2-7b",
+    "aiplanet/effi-7b",
+    "arver/llama7b-qlora",
+    "titan087/OpenLlama13B-Guanaco",
+    "chavinlo/alpaca-native",
+    "project-baize/baize-healthcare-lora-7B",
+    "AlpinDale/pygmalion-instruct",
+    "openlm-research/open_llama_13b",
+    "jondurbin/airoboros-7b-gpt4-1.3",
+    "elyza/ELYZA-japanese-Llama-2-7b-fast",
+    "jondurbin/airoboros-gpt-3.5-turbo-100k-7b",
+    "uukuguy/speechless-codellama-orca-13b",
+    "bigcode/starcoderplus",
+    "TheBloke/guanaco-7B-HF",
+    "Neko-Institute-of-Science/metharme-7b",
+    "TigerResearch/tigerbot-7b-base",
+    "golaxy/gogpt-7b",
+    "togethercomputer/LLaMA-2-7B-32K",
+    "yhyhy3/open_llama_7b_v2_med_instruct",
+    "ajibawa-2023/carl-7b",
+    "stabilityai/stablelm-base-alpha-7b-v2",
+    "conceptofmind/LLongMA-2-7b-16k",
+    "TehVenom/Pygmalion_AlpacaLora-7b",
+    "jondurbin/airoboros-7b-gpt4-1.4.1-qlora",
+    "wannaphong/openthaigpt-0.1.0-beta-full-model_for_open_llm_leaderboard",
+    "ausboss/llama7b-wizardlm-unfiltered",
+    "project-baize/baize-v2-7b",
+    "LMFlow/Robin-v2",
+    "HanningZhang/Robin-v2",
+    "LMFlow/Robin-7b-v2",
+    "OptimalScale/robin-7b-v2-delta",
+    "uukuguy/speechless-codellama-platypus-13b",
+    "jerryjalapeno/nart-100k-7b",
+    "wenge-research/yayi-13b-llama2",
+    "fireballoon/baichuan-vicuna-chinese-7b",
+    "jlevin/guanaco-unchained-llama-2-7b",
+    "csitfun/llama-7b-logicot",
+    "DevaMalla/llama7b_alpaca_1gpu_bf16",
+    "WeOpenML/PandaLM-Alpaca-7B-v1",
+    "illuin/test-custom-llama",
+    "yeontaek/WizardCoder-Python-13B-LoRa",
+    "ashercn97/giraffe-7b",
+    "mosaicml/mpt-7b-chat",
+    "abhishek/autotrain-llama-alpaca-peft-52508123785",
+    "Neko-Institute-of-Science/pygmalion-7b",
+    "TFLai/llama-7b-4bit-alpaca",
+    "huggingface/llama-7b",
+    "TheBloke/Planner-7B-fp16",
+    "shibing624/chinese-llama-plus-13b-hf",
+    "AGI-inc/lora_moe_7b_baseline",
+    "DevaMalla/llama-base-7b",
+    "AGI-inc/lora_moe_7b",
+    "togethercomputer/GPT-JT-6B-v0",
+    "ehartford/WizardLM-7B-Uncensored",
+    "shibing624/chinese-alpaca-plus-7b-hf",
+    "beomi/llama-2-ko-7b",
+    "mosaicml/mpt-7b-8k-instruct",
+    "Enno-Ai/ennodata-7b",
+    "mosaicml/mpt-7b-instruct",
+    "facebook/opt-iml-max-30b",
+    "WeOpenML/Alpaca-7B-v1",
+    "TheBloke/Project-Baize-v2-7B-GPTQ",
+    "codellama/CodeLlama-13b-Instruct-hf",
+    "TheBloke/CodeLlama-13B-Instruct-fp16",
+    "facebook/galactica-30b",
+    "FreedomIntelligence/phoenix-inst-chat-7b",
+    "openlm-research/open_llama_7b_v2",
+    "GeorgiaTechResearchInstitute/galpaca-30b",
+    "THUDM/chatglm2-6b",
+    "togethercomputer/GPT-JT-6B-v1",
+    "TheBloke/koala-7B-HF",
+    "nathan0/mpt_delta_tuned_model_v3",
+    "nathan0/mpt_delta_tuned_model_v2",
+    "GeorgiaTechResearchInstitute/galpaca-30b",
+    "JosephusCheung/Guanaco",
+    "shareAI/CodeLLaMA-chat-13b-Chinese",
+    "TigerResearch/tigerbot-7b-sft",
+    "Writer/InstructPalmyra-20b",
+    "OpenAssistant/codellama-13b-oasst-sft-v10",
+    "bigscience/bloomz-7b1-mt",
+    "nathan0/mpt_delta_tuned_model_v3",
+    "VMware/open-llama-7b-open-instruct",
+    "baichuan-inc/Baichuan-7B",
+    "anas-awadalla/mpt-7b",
+    "mosaicml/mpt-7b",
+    "bigscience/bloomz-7b1",
+    "ziqingyang/chinese-llama-2-7b",
+    "OpenAssistant/codellama-13b-oasst-sft-v10",
+    "wenge-research/yayi-7b",
+    "tiiuae/falcon-7b",
+    "togethercomputer/RedPajama-INCITE-Instruct-7B-v0.1",
+    "togethercomputer/RedPajama-INCITE-7B-Instruct",
+    "TheBloke/landmark-attention-llama7b-fp16",
+    "togethercomputer/GPT-JT-Moderation-6B",
+    "h2oai/h2ogpt-gm-oasst1-en-1024-20b",
+    "dvruette/gpt-neox-20b-full-precision",
+    "TehVenom/Moderator-Chan_GPT-JT-6b",
+    "dvruette/oasst-gpt-neox-20b-1000-steps",
+    "AlekseyKorshuk/pygmalion-6b-vicuna-chatml",
+    "facebook/opt-66b",
+    "Salesforce/codegen-16B-nl",
+    "Vmware/open-llama-7b-v2-open-instruct",
+    "mosaicml/mpt-7b-storywriter",
+    "acrastt/Marx-3B-V2",
+    "openlm-research/open_llama_7b",
+    "Fredithefish/ReasonixPajama-3B-HF",
+    "togethercomputer/GPT-NeoXT-Chat-Base-20B",
+    "psmathur/orca_mini_13b",
+    "RWKV/rwkv-raven-14b",
+    "h2oai/h2ogpt-oasst1-512-20b",
+    "acrastt/Marx-3B",
+    "klosax/open_llama_13b_600bt_preview",
+    "synapsoft/Llama-2-7b-hf-flan2022-1.2M",
+    "OpenAssistant/oasst-sft-1-pythia-12b",
+    "golaxy/gogpt-7b-bloom",
+    "Writer/palmyra-large",
+    "psmathur/orca_mini_7b",
+    "dvruette/oasst-pythia-12b-6000-steps",
+    "NousResearch/CodeLlama-13b-hf",
+    "codellama/CodeLlama-13b-hf",
+    "h2oai/h2ogpt-gm-oasst1-multilang-1024-20b",
+    "VMware/open-llama-0.7T-7B-open-instruct-v1.1",
+    "dvruette/oasst-pythia-12b-flash-attn-5000-steps",
+    "dvruette/oasst-gpt-neox-20b-3000-steps",
+    "RobbeD/OpenLlama-Platypus-3B",
+    "facebook/opt-30b",
+    "acrastt/Puma-3B",
+    "OpenAssistant/oasst-sft-4-pythia-12b-epoch-3.5",
+    "dvruette/oasst-pythia-12b-pretrained-sft",
+    "digitous/GPT-R",
+    "acrastt/Griffin-3B",
+    "togethercomputer/RedPajama-INCITE-Base-7B-v0.1",
+    "togethercomputer/RedPajama-INCITE-7B-Base",
+    "CobraMamba/mamba-gpt-3b-v3",
+    "Danielbrdz/CodeBarcenas-7b",
+    "l3utterfly/open-llama-3b-v2-layla",
+    "CobraMamba/mamba-gpt-3b-v2",
+    "OpenAssistant/pythia-12b-sft-v8-7k-steps",
+    "KoboldAI/GPT-NeoX-20B-Erebus",
+    "RobbeD/Orca-Platypus-3B",
+    "h2oai/h2ogpt-gm-oasst1-en-1024-12b",
+    "OpenAssistant/pythia-12b-sft-v8-2.5k-steps",
+    "AlekseyKorshuk/chatml-pyg-v1",
+    "togethercomputer/RedPajama-INCITE-Chat-7B-v0.1",
+    "togethercomputer/RedPajama-INCITE-7B-Chat",
+    "digitous/Javelin-R",
+    "dvruette/oasst-pythia-12b-reference",
+    "EleutherAI/gpt-neox-20b",
+    "KoboldAI/fairseq-dense-13B",
+    "OpenAssistant/pythia-12b-sft-v8-rlhf-2k-steps",
+    "codellama/CodeLlama-7b-Instruct-hf",
+    "digitous/Javelin-GPTJ",
+    "KoboldAI/GPT-NeoX-20B-Skein",
+    "digitous/Javalion-R",
+    "h2oai/h2ogpt-oasst1-512-12b",
+    "acrastt/Bean-3B",
+    "KoboldAI/GPT-J-6B-Skein",
+    "nomic-ai/gpt4all-j",
+    "databricks/dolly-v2-12b",
+    "TehVenom/Dolly_Shygmalion-6b-Dev_V8P2",
+    "databricks/dolly-v2-7b",
+    "Aspik101/WizardVicuna-Uncensored-3B-instruct-PL-lora_unload",
+    "digitous/Adventien-GPTJ",
+    "openlm-research/open_llama_3b_v2",
+    "RWKV/rwkv-4-14b-pile",
+    "Lazycuber/Janemalion-6B",
+    "OpenAssistant/pythia-12b-pre-v8-12.5k-steps",
+    "digitous/Janin-R",
+    "kfkas/Llama-2-ko-7b-Chat",
+    "heegyu/WizardVicuna-Uncensored-3B-0719",
+    "h2oai/h2ogpt-gm-oasst1-en-1024-open-llama-7b-preview-400bt",
+    "TaylorAI/Flash-Llama-3B",
+    "kfkas/Llama-2-ko-7b-Chat",
+    "digitous/Skegma-GPTJ",
+    "digitous/Javalion-GPTJ",
+    "Pirr/pythia-13b-deduped-green_devil",
+    "TehVenom/PPO_Shygmalion-V8p4_Dev-6b",
+    "dvruette/oasst-pythia-6.9b-4000-steps",
+    "heegyu/WizardVicuna-3B-0719",
+    "psmathur/orca_mini_3b",
+    "OpenAssistant/galactica-6.7b-finetuned",
+    "frank098/orca_mini_3b_juniper",
+    "PygmalionAI/pygmalion-6b",
+    "TehVenom/PPO_Pygway-V8p4_Dev-6b",
+    "TFLai/gpt-neox-20b-4bit-alpaca",
+    "Corianas/gpt-j-6B-Dolly",
+    "TehVenom/Dolly_Shygmalion-6b",
+    "digitous/Janin-GPTJ",
+    "TehVenom/GPT-J-Pyg_PPO-6B-Dev-V8p4",
+    "EleutherAI/gpt-j-6b",
+    "KoboldAI/GPT-J-6B-Shinen",
+    "TehVenom/Dolly_Malion-6b",
+    "TehVenom/ChanMalion",
+    "Salesforce/codegen-6B-nl",
+    "Fredithefish/RedPajama-INCITE-Chat-3B-Instruction-Tuning-with-GPT-4",
+    "KoboldAI/GPT-J-6B-Janeway",
+    "togethercomputer/RedPajama-INCITE-Chat-3B-v1",
+    "togethercomputer/Pythia-Chat-Base-7B",
+    "heegyu/RedTulu-Uncensored-3B-0719",
+    "KoboldAI/PPO_Pygway-6b-Mix",
+    "KoboldAI/OPT-13B-Erebus",
+    "KoboldAI/fairseq-dense-6.7B",
+    "EleutherAI/pythia-12b-deduped",
+    "pszemraj/pythia-6.9b-HC3",
+    "Fredithefish/Guanaco-3B-Uncensored-v2",
+    "facebook/opt-13b",
+    "TehVenom/GPT-J-Pyg_PPO-6B",
+    "EleutherAI/pythia-6.9b-deduped",
+    "Devio/test-1400",
+    "Fredithefish/Guanaco-3B-Uncensored",
+    "codellama/CodeLlama-7b-hf",
+    "acrastt/RedPajama-INCITE-Chat-Instruct-3B-V1",
+    "Fredithefish/ScarletPajama-3B-HF",
+    "KoboldAI/OPT-13B-Nerybus-Mix",
+    "YeungNLP/firefly-bloom-7b1",
+    "DanielSc4/RedPajama-INCITE-Chat-3B-v1-RL-LoRA-8bit-test1",
+    "klosax/open_llama_7b_400bt_preview",
+    "KoboldAI/OPT-13B-Nerys-v2",
+    "TehVenom/PPO_Shygmalion-6b",
+    "amazon/LightGPT",
+    "KnutJaegersberg/black_goo_recipe_c",
+    "NousResearch/CodeLlama-7b-hf",
+    "togethercomputer/RedPajama-INCITE-Instruct-3B-v1",
+    "heegyu/WizardVicuna-open-llama-3b-v2",
+    "bigscience/bloom-7b1",
+    "Devio/test-22B",
+    "RWKV/rwkv-raven-7b",
+    "hakurei/instruct-12b",
+    "CobraMamba/mamba-gpt-3b",
+    "KnutJaegersberg/black_goo_recipe_a",
+    "acrastt/OmegLLaMA-3B",
+    "codellama/CodeLlama-7b-Instruct-hf",
+    "h2oai/h2ogpt-oig-oasst1-512-6_9b",
+    "KoboldAI/OPT-6.7B-Erebus",
+    "facebook/opt-6.7b",
+    "KnutJaegersberg/black_goo_recipe_d",
+    "KnutJaegersberg/LLongMA-3b-LIMA",
+    "KnutJaegersberg/black_goo_recipe_b",
+    "KoboldAI/OPT-6.7B-Nerybus-Mix",
+    "health360/Healix-3B",
+    "EleutherAI/pythia-12b",
+    "Fredithefish/RedPajama-INCITE-Chat-3B-ShareGPT-11K",
+    "GeorgiaTechResearchInstitute/galactica-6.7b-evol-instruct-70k",
+    "h2oai/h2ogpt-oig-oasst1-256-6_9b",
+    "ikala/bloom-zh-3b-chat",
+    "Taekyoon/llama2-ko-7b-test",
+    "anhnv125/pygmalion-6b-roleplay",
+    "TehVenom/DiffMerge_Pygmalion_Main-onto-V8P4",
+    "KoboldAI/OPT-6B-nerys-v2",
+    "Lazycuber/pyg-instruct-wizardlm",
+    "Devio/testC",
+    "KoboldAI/OPT-30B-Erebus",
+    "Fredithefish/CrimsonPajama",
+    "togethercomputer/RedPajama-INCITE-Base-3B-v1",
+    "bigscience/bloomz-3b",
+    "conceptofmind/Open-LLongMA-3b",
+    "RWKV/rwkv-4-7b-pile",
+    "openlm-research/open_llama_3b",
+    "ewof/koishi-instruct-3b",
+    "DanielSc4/RedPajama-INCITE-Chat-3B-v1-FT-LoRA-8bit-test1",
+    "cerebras/Cerebras-GPT-13B",
+    "EleutherAI/pythia-6.7b",
+    "aisquared/chopt-2_7b",
+    "Azure99/blossom-v1-3b",
+    "PSanni/Deer-3b",
+    "bertin-project/bertin-gpt-j-6B-alpaca",
+    "OpenBuddy/openbuddy-openllama-3b-v10-bf16",
+    "KoboldAI/fairseq-dense-2.7B",
+    "ehartford/CodeLlama-34b-Instruct-hf",
+    "codellama/CodeLlama-34b-Instruct-hf",
+    "TheBloke/CodeLlama-34B-Instruct-fp16",
+    "h2oai/h2ogpt-gm-oasst1-en-2048-open-llama-7b-preview-300bt-v2",
+    "openlm-research/open_llama_7b_700bt_preview",
+    "NbAiLab/nb-gpt-j-6B-alpaca",
+    "KoboldAI/OPT-2.7B-Erebus",
+    "Writer/camel-5b-hf",
+    "EleutherAI/pythia-2.7b",
+    "facebook/xglm-7.5B",
+    "EleutherAI/pythia-2.8b-deduped",
+    "klosax/open_llama_3b_350bt_preview",
+    "klosax/openllama-3b-350bt",
+    "KoboldAI/OPT-2.7B-Nerybus-Mix",
+    "KoboldAI/GPT-J-6B-Adventure",
+    "cerebras/Cerebras-GPT-6.7B",
+    "TFLai/pythia-2.8b-4bit-alpaca",
+    "facebook/opt-2.7b",
+    "KoboldAI/OPT-2.7B-Nerys-v2",
+    "bigscience/bloom-3b",
+    "Devio/test100",
+    "RWKV/rwkv-raven-3b",
+    "Azure99/blossom-v2-3b",
+    "codellama/CodeLlama-34b-Python-hf",
+    "bhenrym14/airoboros-33b-gpt4-1.4.1-PI-8192-fp16",
+    "EleutherAI/gpt-neo-2.7B",
+    "danielhanchen/open_llama_3b_600bt_preview",
+    "HuggingFaceH4/starchat-alpha",
+    "pythainlp/wangchanglm-7.5B-sft-en-sharded",
+    "beaugogh/pythia-1.4b-deduped-sharegpt",
+    "HWERI/pythia-1.4b-deduped-sharegpt",
+    "OpenAssistant/stablelm-7b-sft-v7-epoch-3",
+    "codellama/CodeLlama-7b-Python-hf",
+    "aisquared/chopt-1_3b",
+    "PygmalionAI/metharme-1.3b",
+    "Linly-AI/Chinese-LLaMA-2-13B-hf",
+    "chargoddard/llama-2-34b-uncode",
+    "RWKV/rwkv-4-3b-pile",
+    "pythainlp/wangchanglm-7.5B-sft-enth",
+    "MBZUAI/LaMini-GPT-1.5B",
+    "Writer/palmyra-base",
+    "KoboldAI/fairseq-dense-1.3B",
+    "EleutherAI/pythia-1.4b-deduped",
+    "MBZUAI/lamini-neo-1.3b",
+    "h2oai/h2ogpt-gm-oasst1-en-2048-open-llama-7b-preview-300bt",
+    "sartmis1/starcoder-finetune-openapi",
+    "MayaPH/opt-flan-iml-6.7b",
+    "facebook/xglm-4.5B",
+    "WizardLM/WizardCoder-15B-V1.0",
+    "facebook/opt-iml-max-1.3b",
+    "stabilityai/stablelm-tuned-alpha-7b",
+    "aisquared/dlite-v2-1_5b",
+    "stabilityai/stablelm-base-alpha-7b",
+    "sartmis1/starcoder-finetune-selfinstruct",
+    "lizhuang144/starcoder_mirror",
+    "bigcode/starcoder",
+    "TheBloke/CodeLlama-34B-Python-fp16",
+    "open-llm-leaderboard/bloomz-1b7-4bit-alpaca-auto-eval-adapter-applied",
+    "ehartford/CodeLlama-34b-Python-hf",
+    "codellama/CodeLlama-7b-Python-hf",
+    "GeorgiaTechResearchInstitute/starcoder-gpteacher-code-instruct",
+    "LoupGarou/WizardCoder-Guanaco-15B-V1.0",
+    "golaxy/gogpt-3b-bloom",
+    "EleutherAI/pythia-1.3b",
+    "codellama/CodeLlama-13b-Python-hf",
+    "hakurei/lotus-12B",
+    "NYTK/PULI-GPTrio",
+    "facebook/opt-1.3b",
+    "TheBloke/CodeLlama-13B-Python-fp16",
+    "codellama/CodeLlama-13b-Python-hf",
+    "RWKV/rwkv-raven-1b5",
+    "PygmalionAI/pygmalion-2.7b",
+    "bigscience/bloom-1b7",
+    "gpt2-xl",
+    "LoupGarou/WizardCoder-Guanaco-15B-V1.1",
+    "RWKV/rwkv-4-1b5-pile",
+    "codellama/CodeLlama-34b-hf",
+    "NousResearch/CodeLlama-34b-hf",
+    "rinna/bilingual-gpt-neox-4b-8k",
+    "lxe/Cerebras-GPT-2.7B-Alpaca-SP",
+    "cerebras/Cerebras-GPT-2.7B",
+    "jzjiao/opt-1.3b-rlhf",
+    "EleutherAI/gpt-neo-1.3B",
+    "aisquared/dlite-v1-1_5b",
+    "Corianas/Quokka_2.7b",
+    "MrNJK/gpt2-xl-sft",
+    "facebook/galactica-1.3b",
+    "aisquared/dlite-v2-774m",
+    "EleutherAI/pythia-1b-deduped",
+    "Kunhao/pile-7b-250b-tokens",
+    "w601sxs/b1ade-1b",
+    "rinna/bilingual-gpt-neox-4b",
+    "shaohang/SparseOPT-1.3B",
+    "shaohang/Sparse0.5_OPT-1.3",
+    "EleutherAI/polyglot-ko-12.8b",
+    "Salesforce/codegen-6B-multi",
+    "bigscience/bloom-1b1",
+    "TFLai/gpt-neo-1.3B-4bit-alpaca",
+    "FabbriSimo01/Bloom_1b_Quantized",
+    "MBZUAI/LaMini-GPT-774M",
+    "Locutusque/gpt2-large-conversational",
+    "Devio/test-3b",
+    "stabilityai/stablelm-tuned-alpha-3b",
+    "PygmalionAI/pygmalion-1.3b",
+    "KoboldAI/fairseq-dense-355M",
+    "Rachneet/gpt2-xl-alpaca",
+    "gpt2-large",
+    "Mikivis/gpt2-large-lora-sft",
+    "stabilityai/stablelm-base-alpha-3b",
+    "gpt2-medium",
+    "Kunhao/pile-7b",
+    "aisquared/dlite-v1-774m",
+    "aisquared/dlite-v2-355m",
+    "YeungNLP/firefly-bloom-2b6-v2",
+    "KnutJaegersberg/gpt-2-xl-EvolInstruct",
+    "KnutJaegersberg/galactica-orca-wizardlm-1.3b",
+    "cerebras/Cerebras-GPT-1.3B",
+    "FabbriSimo01/Cerebras_1.3b_Quantized",
+    "facebook/xglm-1.7B",
+    "EleutherAI/pythia-410m-deduped",
+    "TheBloke/GPlatty-30B-SuperHOT-8K-fp16",
+    "DataLinguistic/DataLinguistic-34B-V1.0",
+    "Corianas/Quokka_1.3b",
+    "TheTravellingEngineer/bloom-560m-RLHF-v2",
+    "Corianas/1.3b",
+    "RWKV/rwkv-4-430m-pile",
+    "porkorbeef/Llama-2-13b-sf",
+    "xhyi/PT_GPTNEO350_ATG",
+    "TheBloke/Wizard-Vicuna-13B-Uncensored-GPTQ",
+    "bigscience/bloomz-560m",
+    "TheBloke/medalpaca-13B-GPTQ-4bit",
+    "TheBloke/Vicuna-33B-1-3-SuperHOT-8K-fp16",
+    "aisquared/dlite-v1-355m",
+    "uukuguy/speechless-codellama-orca-airoboros-13b-0.10e",
+    "yhyhy3/med-orca-instruct-33b",
+    "TheBloke/Wizard-Vicuna-30B-Superhot-8K-fp16",
+    "TheTravellingEngineer/bloom-1b1-RLHF",
+    "MBZUAI/lamini-cerebras-1.3b",
+    "IDEA-CCNL/Ziya-LLaMA-13B-Pretrain-v1",
+    "TheBloke/WizardLM-7B-uncensored-GPTQ",
+    "TheBloke/EverythingLM-13B-16K-GPTQ",
+    "quantumaikr/open_llama_7b_hf",
+    "TheBloke/chronos-wizardlm-uc-scot-st-13B-GPTQ",
+    "TheBloke/WizardLM-30B-Uncensored-GPTQ",
+    "IDEA-CCNL/Ziya-LLaMA-13B-v1",
+    "Phind/Phind-CodeLlama-34B-v1",
+    "robowaifudev/megatron-gpt2-345m",
+    "MayaPH/GodziLLa-30B-instruct",
+    "TheBloke/CAMEL-33B-Combined-Data-SuperHOT-8K-fp16",
+    "uukuguy/speechless-codellama-orca-platypus-13b-0.10e",
+    "doas/test2",
+    "BreadAi/PM_modelV2",
+    "bigcode/santacoder",
+    "TheBloke/wizard-vicuna-13B-GPTQ",
+    "porkorbeef/Llama-2-13b",
+    "TehVenom/DiffMerge-DollyGPT-Pygmalion",
+    "PygmalionAI/pygmalion-350m",
+    "TheBloke/orca_mini_v3_7B-GPTQ",
+    "TheBloke/WizardLM-Uncensored-SuperCOT-StoryTelling-30B-GPTQ",
+    "TheBloke/WizardLM-30B-GPTQ",
+    "bigscience/bloom-560m",
+    "TFLai/gpt2-turkish-uncased",
+    "TheBloke/guanaco-33B-GPTQ",
+    "TheBloke/openchat_v2_openorca_preview-GPTQ",
+    "porkorbeef/Llama-2-13b-public",
+    "TheBloke/LongChat-13B-GPTQ",
+    "yhyhy3/med-orca-instruct-33b",
+    "TheBloke/airoboros-33B-gpt4-1-4-SuperHOT-8K-fp16",
+    "TheBloke/Chinese-Alpaca-33B-SuperHOT-8K-fp16",
+    "MayaPH/FinOPT-Franklin",
+    "TheBloke/WizardLM-33B-V1.0-Uncensored-GPTQ",
+    "TheBloke/Project-Baize-v2-13B-GPTQ",
+    "malhajar/Platypus2-70B-instruct-4bit-gptq",
+    "KoboldAI/OPT-350M-Erebus",
+    "rishiraj/bloom-560m-guanaco",
+    "Panchovix/WizardLM-33B-V1.0-Uncensored-SuperHOT-8k",
+    "doas/test5",
+    "vicgalle/alpaca-7b",
+    "beomi/KoAlpaca-Polyglot-5.8B",
+    "Phind/Phind-CodeLlama-34B-Python-v1",
+    "timdettmers/guanaco-65b-merged",
+    "TheBloke/wizard-mega-13B-GPTQ",
+    "MayaPH/GodziLLa-30B-plus",
+    "TheBloke/Platypus-30B-SuperHOT-8K-fp16",
+    "facebook/opt-350m",
+    "KoboldAI/OPT-350M-Nerys-v2",
+    "TheBloke/robin-33B-v2-GPTQ",
+    "jaspercatapang/Echidna-30B",
+    "TheBloke/llama-30b-supercot-SuperHOT-8K-fp16",
+    "marcchew/test1",
+    "Harshvir/LaMini-Neo-1.3B-Mental-Health_lora",
+    "golaxy/gogpt-560m",
+    "TheBloke/orca_mini_13B-GPTQ",
+    "Panchovix/airoboros-33b-gpt4-1.2-SuperHOT-8k",
+    "Aspik101/tulu-7b-instruct-pl-lora_unload",
+    "Phind/Phind-CodeLlama-34B-v2",
+    "BreadAi/MusePy-1-2",
+    "cerebras/Cerebras-GPT-590M",
+    "microsoft/CodeGPT-small-py",
+    "victor123/WizardLM-13B-1.0",
+    "OptimalScale/robin-65b-v2-delta",
+    "voidful/changpt-bart",
+    "FabbriSimo01/GPT_Large_Quantized",
+    "MayaPH/FinOPT-Lincoln",
+    "KoboldAI/fairseq-dense-125M",
+    "SebastianSchramm/Cerebras-GPT-111M-instruction",
+    "TheTravellingEngineer/bloom-560m-RLHF",
+    "breadlicker45/dough-instruct-base-001",
+    "WizardLM/WizardLM-30B-V1.0",
+    "WizardLM/WizardLM-30B-V1.0",
+    "WizardLM/WizardLM-30B-V1.0",
+    "TaylorAI/Flash-Llama-30M-20001",
+    "porkorbeef/Llama-2-13b-12_153950",
+    "huggingtweets/bladeecity-jerma985",
+    "KnutJaegersberg/megatron-GPT-2-345m-EvolInstruct",
+    "bhenrym14/airoboros-33b-gpt4-1.4.1-lxctx-PI-16384-fp16",
+    "microsoft/DialoGPT-small",
+    "Corianas/590m",
+    "facebook/xglm-564M",
+    "EleutherAI/gpt-neo-125m",
+    "EleutherAI/pythia-160m-deduped",
+    "klosax/pythia-160m-deduped-step92k-193bt",
+    "MBZUAI/lamini-neo-125m",
+    "bigcode/tiny_starcoder_py",
+    "concedo/OPT-19M-ChatSalad",
+    "anton-l/gpt-j-tiny-random",
+    "grantprice/Cerebras-GPT-590M-finetuned-DND",
+    "deepnight-research/zsc-text",
+    "WangZeJun/bloom-820m-chat",
+    "cerebras/Cerebras-GPT-256M",
+    "ai-forever/rugpt3large_based_on_gpt2",
+    "alibidaran/medical_transcription_generator",
+    "Deci/DeciCoder-1b",
+    "microsoft/DialoGPT-medium",
+    "ogimgio/gpt-neo-125m-neurallinguisticpioneers",
+    "open-llm-leaderboard/bloom-560m-4bit-alpaca-auto-eval-adapter-applied",
+    "BreadAi/gpt-YA-1-1_160M",
+    "microsoft/DialoGPT-large",
+    "facebook/opt-125m",
+    "huggingtweets/jerma985",
+    "Locutusque/gpt2-conversational-or-qa",
+    "concedo/Pythia-70M-ChatSalad",
+    "roneneldan/TinyStories-1M",
+    "BreadAi/DiscordPy",
+    "bigcode/gpt_bigcode-santacoder",
+    "Tincando/fiction_story_generator",
+    "klosax/pythia-70m-deduped-step44k-92bt",
+    "Quake24/easyTermsSummerizer",
+    "BreadAi/gpt-YA-1-1_70M",
+    "EleutherAI/pythia-160m",
+    "euclaise/gpt-neox-122m-minipile-digits",
+    "MBZUAI/lamini-cerebras-590m",
+    "nicholasKluge/Aira-124M",
+    "MayaPH/FinOPT-Washington",
+    "cyberagent/open-calm-large",
+    "BreadAi/StoryPy",
+    "EleutherAI/pythia-70m",
+    "BreadAi/gpt-Youtube",
+    "roneneldan/TinyStories-33M",
+    "EleutherAI/pythia-70m-deduped",
+    "lgaalves/gpt2_guanaco-dolly-platypus",
+    "Corianas/Quokka_590m",
+    "lgaalves/gpt2_platypus-dolly-guanaco",
+    "cyberagent/open-calm-7b",
+    "RWKV/rwkv-4-169m-pile",
+    "gpt2",
+    "roneneldan/TinyStories-28M",
+    "lgaalves/gpt2_open-platypus",
+    "gpt2",
+    "SaylorTwift/gpt2_test",
+    "roneneldan/TinyStories-3M",
+    "nthngdy/pythia-owt2-70m-50k",
+    "Corianas/256_5epoch",
+    "roneneldan/TinyStories-8M",
+    "lgaalves/gpt2-dolly",
+    "nthngdy/pythia-owt2-70m-100k",
+    "aisquared/dlite-v2-124m",
+    "mncai/SGPT-1.3B-insurance-epoch10",
+    "huggingtweets/gladosystem",
+    "abhiramtirumala/DialoGPT-sarcastic-medium",
+    "MBZUAI/lamini-cerebras-256m",
+    "cerebras/Cerebras-GPT-111M",
+    "uberkie/metharme-1.3b-finetuned",
+    "MBZUAI/lamini-cerebras-111m",
+    "psyche/kogpt",
+    "Corianas/Quokka_256m",
+    "vicgalle/gpt2-alpaca-gpt4",
+    "aisquared/dlite-v1-124m",
+    "Mikivis/xuanxuan",
+    "MBZUAI/LaMini-GPT-124M",
+    "vicgalle/gpt2-alpaca",
+    "huashiyiqike/testmodel",
+    "Corianas/111m",
+    "baseline",
+]

src/tools/plots.py

@@ -0,0 +1,154 @@
+import pandas as pd
+import numpy as np
+import plotly.express as px
+from plotly.graph_objs import Figure
+
+from src.leaderboard.filter_models import FLAGGED_MODELS
+from src.display.utils import human_baseline_row as HUMAN_BASELINE, AutoEvalColumn, Tasks, Task, BENCHMARK_COLS
+from src.leaderboard.read_evals import EvalResult
+
+
+
+def create_scores_df(raw_data: list[EvalResult]) -> pd.DataFrame:
+    """
+    Generates a DataFrame containing the maximum scores until each date.
+
+    :param results_df: A DataFrame containing result information including metric scores and dates.
+    :return: A new DataFrame containing the maximum scores until each date for every metric.
+    """
+    # Step 1: Ensure 'date' is in datetime format and sort the DataFrame by it
+    results_df = pd.DataFrame(raw_data)
+    #results_df["date"] = pd.to_datetime(results_df["date"], format="mixed", utc=True)
+    results_df.sort_values(by="date", inplace=True)
+
+    # Step 2: Initialize the scores dictionary
+    scores = {k: [] for k in BENCHMARK_COLS + [AutoEvalColumn.average.name]}
+
+    # Step 3: Iterate over the rows of the DataFrame and update the scores dictionary
+    for task in [t.value for t in Tasks] + [Task("Average", "avg", AutoEvalColumn.average.name)]:
+        current_max = 0
+        last_date = ""
+        column = task.col_name
+        for _, row in results_df.iterrows():
+            current_model = row["full_model"]
+            if current_model in FLAGGED_MODELS:
+                continue
+
+            current_date = row["date"]
+            if task.benchmark == "Average":
+                current_score = np.mean(list(row["results"].values()))
+            else:
+                current_score = row["results"][task.benchmark]
+
+            if current_score > current_max:
+                if current_date == last_date and len(scores[column]) > 0:
+                    scores[column][-1] = {"model": current_model, "date": current_date, "score": current_score}
+                else:
+                    scores[column].append({"model": current_model, "date": current_date, "score": current_score})
+                current_max = current_score
+                last_date = current_date
+
+    # Step 4: Return all dictionaries as DataFrames
+    return {k: pd.DataFrame(v) for k, v in scores.items()}
+
+
+def create_plot_df(scores_df: dict[str: pd.DataFrame]) -> pd.DataFrame:
+    """
+    Transforms the scores DataFrame into a new format suitable for plotting.
+
+    :param scores_df: A DataFrame containing metric scores and dates.
+    :return: A new DataFrame reshaped for plotting purposes.
+    """
+    # Initialize the list to store DataFrames
+    dfs = []
+
+    # Iterate over the cols and create a new DataFrame for each column
+    for col in BENCHMARK_COLS + [AutoEvalColumn.average.name]:
+        d = scores_df[col].reset_index(drop=True)
+        d["task"] = col
+        dfs.append(d)
+
+    # Concatenate all the created DataFrames
+    concat_df = pd.concat(dfs, ignore_index=True)
+
+    # Sort values by 'date'
+    concat_df.sort_values(by="date", inplace=True)
+    concat_df.reset_index(drop=True, inplace=True)
+    return concat_df
+
+
+def create_metric_plot_obj(
+    df: pd.DataFrame, metrics: list[str], title: str
+) -> Figure:
+    """
+    Create a Plotly figure object with lines representing different metrics
+    and horizontal dotted lines representing human baselines.
+
+    :param df: The DataFrame containing the metric values, names, and dates.
+    :param metrics: A list of strings representing the names of the metrics
+                    to be included in the plot.
+    :param title: A string representing the title of the plot.
+    :return: A Plotly figure object with lines representing metrics and
+             horizontal dotted lines representing human baselines.
+    """
+
+    # Filter the DataFrame based on the specified metrics
+    df = df[df["task"].isin(metrics)]
+
+    # Filter the human baselines based on the specified metrics
+    filtered_human_baselines = {k: v for k, v in HUMAN_BASELINE.items() if k in metrics}
+
+    # Create a line figure using plotly express with specified markers and custom data
+    fig = px.line(
+        df,
+        x="date",
+        y="score",
+        color="task",
+        markers=True,
+        custom_data=["task", "score", "model"],
+        title=title,
+    )
+
+    # Update hovertemplate for better hover interaction experience
+    fig.update_traces(
+        hovertemplate="<br>".join(
+            [
+                "Model Name: %{customdata[2]}",
+                "Metric Name: %{customdata[0]}",
+                "Date: %{x}",
+                "Metric Value: %{y}",
+            ]
+        )
+    )
+
+    # Update the range of the y-axis
+    fig.update_layout(yaxis_range=[0, 100])
+
+    # Create a dictionary to hold the color mapping for each metric
+    metric_color_mapping = {}
+
+    # Map each metric name to its color in the figure
+    for trace in fig.data:
+        metric_color_mapping[trace.name] = trace.line.color
+
+    # Iterate over filtered human baselines and add horizontal lines to the figure
+    for metric, value in filtered_human_baselines.items():
+        color = metric_color_mapping.get(metric, "blue")  # Retrieve color from mapping; default to blue if not found
+        location = "top left" if metric == "HellaSwag" else "bottom left"  # Set annotation position
+        # Add horizontal line with matched color and positioned annotation
+        fig.add_hline(
+            y=value,
+            line_dash="dot",
+            annotation_text=f"{metric} human baseline",
+            annotation_position=location,
+            annotation_font_size=10,
+            annotation_font_color=color,
+            line_color=color,
+        )
+
+    return fig
+
+
+# Example Usage:
+# human_baselines dictionary is defined.
+# chart = create_metric_plot_obj(scores_df, ["ARC", "HellaSwag", "MMLU", "TruthfulQA"], human_baselines, "Graph Title")