app.py

import gradio as gr
import pandas as pd
import matplotlib.pyplot as plt

# Load datasets
leaderboard_data = pd.read_parquet(
    "hf://datasets/alibayram/yapay_zeka_turkce_mmlu_liderlik_tablosu/data/train-00000-of-00001.parquet"
)
model_responses_data = pd.read_parquet(
    "hf://datasets/alibayram/yapay_zeka_turkce_mmlu_model_cevaplari/data/train-00000-of-00001.parquet"
)
section_results_data = pd.read_parquet(
    "hf://datasets/alibayram/yapay_zeka_turkce_mmlu_bolum_sonuclari/data/train-00000-of-00001.parquet"
)

# Leaderboard Tab
def get_leaderboard(sort_by="Accuracy"):
    return leaderboard_data.sort_values(by=sort_by, ascending=False)

# Model Responses Tab
def search_model_responses(query, model):
    filtered = model_responses_data[
        (model_responses_data["model"] == model) &
        (model_responses_data["question"].str.contains(query, case=False))
    ]
    return filtered

# Section Results Tab
def plot_section_results():
    fig, ax = plt.subplots(figsize=(10, 6))
    section_results_data.groupby("section")["accuracy"].mean().plot(kind="bar", ax=ax)
    ax.set_title("Section-Wise Performance")
    ax.set_ylabel("Accuracy (%)")
    ax.set_xlabel("Section")
    return fig

# Model Comparison Tab
def compare_models(models):
    comparison = leaderboard_data[leaderboard_data["model"].isin(models)]
    return comparison

# Gradio Interface
with gr.Blocks() as app:
    gr.Markdown("# 🏆 Turkish MMLU Leaderboard")
    gr.Markdown("Explore the performance of AI models on Turkish MMLU benchmarks.")

    with gr.Tab("Leaderboard"):
        sort_by = gr.Dropdown(
            ["Accuracy", "Runtime", "Model Name"],
            label="Sort By",
            value="Accuracy"
        )
        leaderboard_table = gr.DataFrame(value=leaderboard_data)
        sort_by.change(get_leaderboard, inputs=sort_by, outputs=leaderboard_table)

    with gr.Tab("Model Responses"):
        model_dropdown = gr.Dropdown(
            leaderboard_data["model"].unique(), label="Select Model"
        )
        query_input = gr.Textbox(label="Search Query")
        responses_output = gr.DataFrame()
        query_input.change(
            search_model_responses,
            inputs=[query_input, model_dropdown],
            outputs=responses_output,
        )

    with gr.Tab("Section Results"):
        gr.Markdown("### Section-Wise Results")
        gr.Plot(plot_section_results)

    with gr.Tab("Model Comparison"):
        model_select = gr.CheckboxGroup(
            options=leaderboard_data["model"].unique(), label="Select Models"
        )
        comparison_table = gr.DataFrame()
        model_select.change(compare_models, inputs=model_select, outputs=comparison_table)

app.launch()