app.py

import io
import json

import gradio as gr
import pandas as pd
from huggingface_hub import HfFileSystem


RESULTS_DATASET_ID = "datasets/open-llm-leaderboard/results"
EXCLUDED_KEYS =  {
    "pretty_env_info",
    "chat_template",
    "group_subtasks",
}
EXCLUDED_RESULTS_KEYS = {
    "leaderboard",
}
EXCLUDED_RESULTS_LEADERBOARDS_KEYS = {
    "alias",
}

DEFAULT_HTML_TABLE = """
<table>
  <thead>
    <tr>
      <th>Parameters</th>
      <th>Model-1</th>
      <th>Model-2</th>
    </tr>
  </thead>
  <tbody>
  </tbody>
</table>
"""


fs = HfFileSystem()


def fetch_result_paths():
    paths = fs.glob(f"{RESULTS_DATASET_ID}/**/**/*.json")
    return paths


def filter_latest_result_path_per_model(paths):
    from collections import defaultdict

    d = defaultdict(list)
    for path in paths:
        model_id, _ = path[len(RESULTS_DATASET_ID) +1:].rsplit("/", 1)
        d[model_id].append(path)
    return {model_id: max(paths) for model_id, paths in d.items()}


def get_result_path_from_model(model_id, result_path_per_model):
    return result_path_per_model[model_id]


def load_data(result_path) -> pd.DataFrame:
    with fs.open(result_path, "r") as f:
        data = json.load(f)
    return data


def load_result(model_id):
    result_path = get_result_path_from_model(model_id, latest_result_path_per_model)
    data = load_data(result_path)
    df = to_dataframe(data)
    result = [
        # to_vertical(df),
        to_vertical(filter_results(df)),
        to_vertical(filter_configs(df)),
    ]
    return result


def to_vertical(df):
    df = df.T.rename_axis("Parameters")
    df.index = df.index.str.join(".")
    return df


def to_dataframe(data):
    df = pd.json_normalize([{key: value for key, value in data.items() if key not in EXCLUDED_KEYS}])
    # df.columns = df.columns.str.split(".")  # .split return a list instead of a tuple
    df.columns = list(map(lambda x: tuple(x.split(".")), df.columns))
    df.index = [data.get("model_name", "Model")]
    return df


def filter_results(df):
    df = df.loc[:, df.columns.str[0] == "results"]
    df = df.loc[:, ~df.columns.str[1].isin(EXCLUDED_RESULTS_KEYS)]
    # df.columns.str[1].str = df.columns.str[1].str.removeprefix("leaderboard_")
    df = df.loc[:, ~df.columns.str[2].isin(EXCLUDED_RESULTS_LEADERBOARDS_KEYS)]
    df.columns = df.columns.str[1:]
    df.columns = map(lambda x: (x[0].removeprefix("leaderboard_"), *x[1:]), df.columns)
    return df


def filter_configs(df):
    df = df.loc[:, df.columns.str[0] == "configs"]
    # df = df.loc[:, ~df.columns.str[1].isin(EXCLUDED_RESULTS_KEYS)]
    # df = df.loc[:, ~df.columns.str[2].isin(EXCLUDED_RESULTS_LEADERBOARDS_KEYS)]
    df.columns = df.columns.str[1:]
    df.columns = map(lambda x: (x[0].removeprefix("leaderboard_"), *x[1:]), df.columns)
    return df


def concat_result_1(result_1, results):
    results = pd.read_html(io.StringIO(results))[0]
    return (
        pd.concat([result_1, results.iloc[:, [0, 2]].set_index("Parameters")], axis=1)
        .reset_index()
        .fillna("")
        .to_html(index=False)
    )


def concat_result_2(result_2, results):
    results = pd.read_html(io.StringIO(results))[0]
    return (
        pd.concat([results.iloc[:, [0, 1]].set_index("Parameters"), result_2], axis=1)
        .reset_index()
        .fillna("")
        .to_html(index=False)
    )


def render_result_1(model_id, *results):
    result = load_result(model_id)
    return [concat_result_1(*result_args) for result_args in zip(result, results)]


def render_result_2(model_id, *results):
    result = load_result(model_id)
    return [concat_result_2(*result_args) for result_args in zip(result, results)]


# if __name__ == "__main__":
latest_result_path_per_model = filter_latest_result_path_per_model(fetch_result_paths())

with gr.Blocks(fill_height=True) as demo:
    gr.HTML("<h1 style='text-align: center;'>Compare Results of the 🤗 Open LLM Leaderboard</h1>")
    gr.HTML("<h3 style='text-align: center;'>Select 2 results to load and compare</h3>")

    with gr.Row():
        with gr.Column():
            model_id_1 = gr.Dropdown(choices=list(latest_result_path_per_model.keys()), label="Results")
            load_btn_1 = gr.Button("Load")
        with gr.Column():
            model_id_2 = gr.Dropdown(choices=list(latest_result_path_per_model.keys()), label="Results")
            load_btn_2 = gr.Button("Load")

    results = []
    with gr.Row():
        # with gr.Tab("All"):
        #     # results.append(gr.Dataframe(
        #     #     label="Results",
        #     #     headers=["Parameters", "Model-1", "Model-2"],
        #     #     interactive=False,
        #     #     column_widths=["30%", "30%", "30%"],
        #     #     wrap=True,
        #     # ))
        #     results.append(gr.HTML(value=DEFAULT_HTML_TABLE))
        with gr.Tab("Results"):
            # results.append(gr.Dataframe(
            #     label="Results",
            #     headers=["Parameters", "Model-1", "Model-2"],
            #     interactive=False,
            #     column_widths=["30%", "30%", "30%"],
            #     wrap=True,
            # ))
            results.append(gr.HTML(value=DEFAULT_HTML_TABLE))
        with gr.Tab("Configs"):
            # results.append(gr.Dataframe(
            #     label="Results",
            #     headers=["Parameters", "Model-1", "Model-2"],
            #     interactive=False,
            #     column_widths=["30%", "30%", "30%"],
            #     wrap=True,
            # ))
            results.append(gr.HTML(value=DEFAULT_HTML_TABLE))

    load_btn_1.click(
        fn=render_result_1,
        inputs=[model_id_1, *results],
        outputs=[*results],
    )
    load_btn_2.click(
        fn=render_result_2,
        inputs=[model_id_2, *results],
        outputs=[*results],
    )

demo.launch()