app.py

import gradio as gr
import pandas as pd

from src.about import (  # CITATION_BUTTON_LABEL,; CITATION_BUTTON_TEXT,; EVALUATION_QUEUE_TEXT,
    CITATION_BUTTON_LABEL,
    CITATION_BUTTON_TEXT,
    INTRODUCTION_TEXT,
    LLM_BENCHMARKS_TEXT,
    TITLE,
)
from src.display.css_html_js import custom_css
from src.display.utils import (  # EVAL_TYPES,; WeightType,; BENCHMARK_COLS,; EVAL_COLS,; NUMERIC_INTERVALS,; ModelType,; Precision,
    COLS,
    TYPES,
    AutoEvalColumn,
    fields,
)

# from src.envs import API, EVAL_REQUESTS_PATH, EVAL_RESULTS_PATH, QUEUE_REPO, REPO_ID, RESULTS_REPO, TOKEN
from src.envs import CRM_RESULTS_PATH
from src.populate import get_leaderboard_df_crm

original_df = get_leaderboard_df_crm(CRM_RESULTS_PATH, COLS)

# raw_data, original_df = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, COLS, BENCHMARK_COLS)
leaderboard_df = original_df.copy()
# leaderboard_df = leaderboard_df.style.format({"accuracy_metric_average": "{0:.2f}"})


# Searching and filtering
def update_table(
    hidden_df: pd.DataFrame,
    columns: list,
    llm_query: list,
    llm_provider_query: list,
    accuracy_method_query: str,
    use_case_query: list,
    use_case_type_query: list,
    # type_query: list,
    # precision_query: str,
    # size_query: list,
    # show_deleted: bool,
    # query: str,
):
    # filtered_df = filter_models(hidden_df, type_query, size_query, precision_query, show_deleted)
    # filtered_df = filter_queries(query, filtered_df)
    filtered_df = filter_llm_func(hidden_df, llm_query)
    filtered_df = filter_llm_provider_func(filtered_df, llm_provider_query)
    filtered_df = filter_accuracy_method_func(filtered_df, accuracy_method_query)

    filtered_df["Use Case Area"] = filtered_df["Use Case Name"].apply(lambda x: x.split(": ")[0])
    # print(filtered_df["Use Case Area"].unique())
    filtered_df = filter_use_case_func(filtered_df, use_case_query)
    filtered_df = filter_use_case_type_func(filtered_df, use_case_type_query)
    df = select_columns(filtered_df, columns)
    return df


def filter_accuracy_method_func(df: pd.DataFrame, accuracy_method_query: str) -> pd.DataFrame:
    return df[df["Accuracy Method"] == accuracy_method_query]


def filter_use_case_func(df: pd.DataFrame, use_case_query: list) -> pd.DataFrame:
    # print(use_case_query)
    # print(df[df["Use Case Name"].isin(["Service: Conversation summary"])])
    return df[df["Use Case Name"].isin(use_case_query)]


def filter_use_case_type_func(df: pd.DataFrame, use_case_type_query: list) -> pd.DataFrame:
    return df[df["Use Case Type"].isin(use_case_type_query)]


def filter_llm_func(df: pd.DataFrame, llm_query: list) -> pd.DataFrame:
    return df[df["Model Name"].isin(llm_query)]


def filter_llm_provider_func(df: pd.DataFrame, llm_provider_query: list) -> pd.DataFrame:
    return df[df["LLM Provider"].isin(llm_provider_query)]


# def search_table(df: pd.DataFrame, query: str) -> pd.DataFrame:
#     return df[(df[AutoEvalColumn.model.name].str.contains(query, case=False))]


def select_columns(df: pd.DataFrame, columns: list) -> pd.DataFrame:
    always_here_cols = [
        AutoEvalColumn.model.name,
    ]
    # We use COLS to maintain sorting
    filtered_df = df[always_here_cols + [c for c in COLS if c in df.columns and c in columns]]
    return filtered_df


# def filter_queries(query: str, filtered_df: pd.DataFrame) -> pd.DataFrame:
#     final_df = []
#     if query != "":
#         queries = [q.strip() for q in query.split(";")]
#         for _q in queries:
#             _q = _q.strip()
#             if _q != "":
#                 temp_filtered_df = search_table(filtered_df, _q)
#                 if len(temp_filtered_df) > 0:
#                     final_df.append(temp_filtered_df)
#         if len(final_df) > 0:
#             filtered_df = pd.concat(final_df)
#             filtered_df = filtered_df.drop_duplicates(
#                 subset=[AutoEvalColumn.model.name, AutoEvalColumn.precision.name, AutoEvalColumn.revision.name]
#             )

#     return filtered_df


# def filter_models(
#     df: pd.DataFrame, type_query: list, size_query: list, precision_query: list, show_deleted: bool
# ) -> pd.DataFrame:
#     # Show all models
#     filtered_df = df
#     # if show_deleted:
#     #     filtered_df = df
#     # else:  # Show only still on the hub models
#     #     filtered_df = df[df[AutoEvalColumn.still_on_hub.name] is True]

#     type_emoji = [t[0] for t in type_query]
#     filtered_df = filtered_df.loc[df[AutoEvalColumn.model_type_symbol.name].isin(type_emoji)]
#     filtered_df = filtered_df.loc[df[AutoEvalColumn.precision.name].isin(precision_query + ["None"])]

#     numeric_interval = pd.IntervalIndex(sorted([NUMERIC_INTERVALS[s] for s in size_query]))
#     params_column = pd.to_numeric(df[AutoEvalColumn.params.name], errors="coerce")
#     mask = params_column.apply(lambda x: any(numeric_interval.contains(x)))
#     filtered_df = filtered_df.loc[mask]

#     return filtered_df


demo = gr.Blocks(css=custom_css)
with demo:
    gr.HTML(TITLE)
    gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")

    with gr.Tabs(elem_classes="tab-buttons") as tabs:
        with gr.TabItem("🏅 LLM Benchmark", elem_id="llm-benchmark-tab-table", id=0):
            with gr.Row():
                with gr.Column():
                    # with gr.Row():
                    #     search_bar = gr.Textbox(
                    #         placeholder=" 🔍 Search for your model (separate multiple queries with `;`) and press ENTER...",
                    #         show_label=False,
                    #         elem_id="search-bar",
                    #     )
                    with gr.Row():
                        shown_columns = gr.CheckboxGroup(
                            choices=[c.name for c in fields(AutoEvalColumn) if not c.hidden and not c.never_hidden],
                            value=[
                                c.name
                                for c in fields(AutoEvalColumn)
                                if c.displayed_by_default and not c.hidden and not c.never_hidden
                            ],
                            label="Select columns to show",
                            elem_id="column-select",
                            interactive=True,
                        )
                # with gr.Column(min_width=320):
                #     # with gr.Box(elem_id="box-filter"):
                #     filter_columns_type = gr.CheckboxGroup(
                #         label="Model types",
                #         choices=[t.to_str() for t in ModelType],
                #         value=[t.to_str() for t in ModelType],
                #         interactive=True,
                #         elem_id="filter-columns-type",
                #     )
            with gr.Row():
                with gr.Column():
                    filter_use_case_type = gr.CheckboxGroup(
                        choices=["Service", "Sales"],
                        value=["Service", "Sales"],
                        label="Use Case Area",
                        info="",
                        interactive=True,
                    )
                with gr.Column():
                    filter_use_case_type = gr.CheckboxGroup(
                        choices=["Summary", "Generation"],
                        value=["Summary", "Generation"],
                        label="Use Case Type",
                        info="",
                        interactive=True,
                    )
                with gr.Column():
                    filter_use_case = gr.Dropdown(
                        choices=list(original_df["Use Case Name"].unique()),
                        value=list(original_df["Use Case Name"].unique()),
                        label="Use Case",
                        info="",
                        multiselect=True,
                        interactive=True,
                    )
                with gr.Column():
                    filter_metric_area = gr.CheckboxGroup(
                        choices=["Accuracy", "Speed (Latency)", "Trust & Safety", "Cost"],
                        value=["Accuracy", "Speed (Latency)", "Trust & Safety", "Cost"],
                        label="Metric Area",
                        info="",
                        interactive=True,
                    )
                with gr.Column():
                    filter_accuracy_method = gr.Radio(
                        choices=["Manual", "Auto"],
                        value="Manual",
                        label="Accuracy Method",
                        info="",
                        interactive=True,
                    )
                with gr.Column():
                    filter_accuracy_threshold = gr.Number(
                        value="3",
                        label="Accuracy Threshold",
                        info="Range: 0.0 to 4.0",
                        interactive=True,
                    )
                with gr.Column():
                    filter_llm = gr.CheckboxGroup(
                        choices=list(original_df["Model Name"].unique()),
                        value=list(leaderboard_df["Model Name"].unique()),
                        label="Model Name",
                        info="",
                        interactive=True,
                    )
                with gr.Column():
                    filter_llm_provider = gr.CheckboxGroup(
                        choices=list(original_df["LLM Provider"].unique()),
                        value=list(leaderboard_df["LLM Provider"].unique()),
                        label="LLM Provider",
                        info="",
                        interactive=True,
                    )

            leaderboard_table = gr.components.Dataframe(
                value=leaderboard_df[[c.name for c in fields(AutoEvalColumn) if c.never_hidden] + shown_columns.value],
                headers=[c.name for c in fields(AutoEvalColumn) if c.never_hidden] + shown_columns.value,
                datatype=TYPES,
                elem_id="leaderboard-table",
                interactive=False,
                visible=True,
            )

            # Dummy leaderboard for handling the case when the user uses backspace key
            hidden_leaderboard_table_for_search = gr.components.Dataframe(
                value=original_df[COLS],
                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,
            #         search_bar,
            #     ],
            #     leaderboard_table,
            # )
            for selector in [
                shown_columns,
                filter_llm,
                filter_llm_provider,
                filter_accuracy_method,
                filter_use_case,
                filter_use_case_type,
                # filter_columns_type,
                # filter_columns_precision,
                # filter_columns_size,
                # deleted_models_visibility,
            ]:
                selector.change(
                    update_table,
                    [
                        hidden_leaderboard_table_for_search,
                        shown_columns,
                        filter_llm,
                        filter_llm_provider,
                        filter_accuracy_method,
                        filter_use_case,
                        filter_use_case_type,
                        # filter_columns_type,
                        # filter_columns_precision,
                        # filter_columns_size,
                        # deleted_models_visibility,
                        # search_bar,
                    ],
                    leaderboard_table,
                    queue=True,
                )

        with gr.TabItem("📝 About", elem_id="llm-benchmark-tab-table", id=2):
            gr.Markdown(LLM_BENCHMARKS_TEXT, elem_classes="markdown-text")

    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.queue(default_concurrency_limit=40).launch()