app.py

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

# Input data
data_full = [
    ["CultriX/Qwen2.5-14B-SLERPv7", 0.7205, 0.8272, 0.7541, 0.6581, 0.5000, 0.7290],
    ["djuna/Q2.5-Veltha-14B-0.5", 0.7492, 0.8386, 0.7305, 0.5980, 0.4300, 0.7817],
    ["CultriX/Qwen2.5-14B-FinalMerge", 0.7248, 0.8277, 0.7113, 0.7052, 0.5700, 0.7001],
    ["CultriX/Qwen2.5-14B-MultiCultyv2", 0.7295, 0.8359, 0.7363, 0.5767, 0.4400, 0.7316],
    ["CultriX/Qwen2.5-14B-Brocav7", 0.7445, 0.8353, 0.7508, 0.6292, 0.4600, 0.7629],
    ["CultriX/Qwen2.5-14B-Broca", 0.7456, 0.8352, 0.7480, 0.6034, 0.4400, 0.7716],
    ["CultriX/Qwen2.5-14B-Brocav3", 0.7395, 0.8388, 0.7393, 0.6405, 0.4700, 0.7659],
    ["CultriX/Qwen2.5-14B-Brocav4", 0.7432, 0.8377, 0.7444, 0.6277, 0.4800, 0.7580],
    ["CultriX/Qwen2.5-14B-Brocav2", 0.7492, 0.8302, 0.7508, 0.6377, 0.5100, 0.7478],
    ["CultriX/Qwen2.5-14B-Brocav5", 0.7445, 0.8313, 0.7547, 0.6376, 0.5000, 0.7304],
    ["CultriX/Qwen2.5-14B-Brocav6", 0.7179, 0.8354, 0.7531, 0.6378, 0.4900, 0.7524],
    ["CultriX/Qwenfinity-2.5-14B", 0.7347, 0.8254, 0.7279, 0.7267, 0.5600, 0.6970],
    ["CultriX/Qwen2.5-14B-Emergedv2", 0.7137, 0.8335, 0.7363, 0.5836, 0.4400, 0.7344],
    ["CultriX/Qwen2.5-14B-Unity", 0.7063, 0.8343, 0.7423, 0.6820, 0.5700, 0.7498],
    ["CultriX/Qwen2.5-14B-MultiCultyv3", 0.7132, 0.8216, 0.7395, 0.6792, 0.5500, 0.7120],
    ["CultriX/Qwen2.5-14B-Emergedv3", 0.7436, 0.8312, 0.7519, 0.6585, 0.5500, 0.7068],
    ["CultriX/SeQwence-14Bv1", 0.7278, 0.8410, 0.7541, 0.6816, 0.5200, 0.7539],
    ["CultriX/Qwen2.5-14B-Wernickev2", 0.7391, 0.8168, 0.7273, 0.6220, 0.4500, 0.7572],
    ["CultriX/Qwen2.5-14B-Wernickev3", 0.7357, 0.8148, 0.7245, 0.7023, 0.5500, 0.7869],
    ["CultriX/Qwen2.5-14B-Wernickev4", 0.7355, 0.8290, 0.7497, 0.6306, 0.4800, 0.7635],
    ["CultriX/SeQwential-14B-v1", 0.7355, 0.8205, 0.7549, 0.6367, 0.4800, 0.7626],
    ["CultriX/Qwen2.5-14B-Wernickev5", 0.7224, 0.8272, 0.7541, 0.6790, 0.5100, 0.7578],
    ["CultriX/Qwen2.5-14B-Wernickev6", 0.6994, 0.7549, 0.5816, 0.6991, 0.5800, 0.7267],
    ["CultriX/Qwen2.5-14B-Wernickev7", 0.7147, 0.7599, 0.6097, 0.7056, 0.5700, 0.7164],
    ["CultriX/Qwen2.5-14B-FinalMerge-tmp2", 0.7255, 0.8192, 0.7535, 0.6671, 0.5000, 0.7612],
]

columns = ["Model Configuration", "tinyArc", "tinyHellaswag", "tinyMMLU", "tinyTruthfulQA", "tinyTruthfulQA_mc1", "tinyWinogrande"]

# Convert to DataFrame
df_full = pd.DataFrame(data_full, columns=columns)


def plot_average_scores():
    df_full["Average Score"] = df_full.iloc[:, 1:].mean(axis=1)
    df_avg_sorted = df_full.sort_values(by="Average Score", ascending=False)

    plt.figure(figsize=(12, 8))
    plt.barh(df_avg_sorted["Model Configuration"], df_avg_sorted["Average Score"])
    plt.title("Average Performance of Models Across Tasks", fontsize=16)
    plt.xlabel("Average Score", fontsize=14)
    plt.ylabel("Model Configuration", fontsize=14)
    plt.gca().invert_yaxis()
    plt.grid(axis='x', linestyle='--', alpha=0.7)
    plt.tight_layout()
    plt.savefig("average_performance.png")
    return "average_performance.png"

def plot_task_performance():
    df_full_melted = df_full.melt(id_vars="Model Configuration", var_name="Task", value_name="Score")

    plt.figure(figsize=(14, 10))
    for model in df_full["Model Configuration"]:
        model_data = df_full_melted[df_full_melted["Model Configuration"] == model]
        plt.plot(model_data["Task"], model_data["Score"], marker="o", label=model)

    plt.title("Performance of All Models Across Tasks", fontsize=16)
    plt.xlabel("Task", fontsize=14)
    plt.ylabel("Score", fontsize=14)
    plt.xticks(rotation=45)
    plt.legend(bbox_to_anchor=(1.05, 1), loc='upper left', fontsize=9)
    plt.grid(axis='y', linestyle='--', alpha=0.7)
    plt.tight_layout()
    plt.savefig("task_performance.png")
    return "task_performance.png"

def plot_task_specific_top_models():
    top_models = df_full.iloc[:, :-1].set_index("Model Configuration").idxmax()
    top_scores = df_full.iloc[:, :-1].set_index("Model Configuration").max()

    results = pd.DataFrame({"Top Model": top_models, "Score": top_scores}).reset_index().rename(columns={"index": "Task"})

    plt.figure(figsize=(12, 6))
    plt.bar(results["Task"], results["Score"])
    plt.title("Task-Specific Top Models", fontsize=16)
    plt.xlabel("Task", fontsize=14)
    plt.ylabel("Score", fontsize=14)
    plt.grid(axis="y", linestyle="--", alpha=0.7)
    plt.tight_layout()
    plt.savefig("task_specific_top_models.png")
    return "task_specific_top_models.png"

def top_3_models_per_task():
    top_3_data = {
        task: df_full.nlargest(3, task)[["Model Configuration", task]].values.tolist()
        for task in df_full.columns[1:-1]
    }
    top_3_results = pd.DataFrame({
        task: {
            "Top 3 Models": [entry[0] for entry in top_3_data[task]],
            "Scores": [entry[1] for entry in top_3_data[task]],
        }
        for task in top_3_data
    }).T.rename_axis("Task").reset_index()
    return top_3_results

def summary_statistics():
    stats = df_full.iloc[:, 1:].describe().T  # Summary stats for each task
    stats['Std Dev'] = df_full.iloc[:, 1:].std(axis=0)
    return stats.reset_index()

def plot_distribution_boxplots():
    plt.figure(figsize=(14, 8))
    df_melted = df_full.melt(id_vars="Model Configuration", var_name="Task", value_name="Score")
    sns.boxplot(x="Task", y="Score", data=df_melted)
    plt.title("Score Distribution by Task", fontsize=16)
    plt.xlabel("Task", fontsize=14)
    plt.ylabel("Score", fontsize=14)
    plt.grid(axis='y', linestyle='--', alpha=0.7)
    plt.tight_layout()
    plt.savefig("distribution_boxplots.png")
    return "distribution_boxplots.png"

def best_overall_model():
    df_full["Average Score"] = df_full.iloc[:, 1:].mean(axis=1)
    best_model = df_full.loc[df_full["Average Score"].idxmax()]
    return best_model

def plot_heatmap():
    plt.figure(figsize=(12, 8))
    sns.heatmap(df_full.iloc[:, 1:], annot=True, cmap="YlGnBu", xticklabels=columns[1:], yticklabels=df_full["Model Configuration"])
    plt.title("Performance Heatmap", fontsize=16)
    plt.tight_layout()
    plt.savefig("performance_heatmap.png")
    return "performance_heatmap.png"

with gr.Blocks() as demo:
    gr.Markdown("# Model Performance Analysis")

    with gr.Row():
        btn1 = gr.Button("Show Average Performance")
        img1 = gr.Image(type="filepath")
        btn1.click(plot_average_scores, outputs=img1)

    with gr.Row():
        btn2 = gr.Button("Show Task Performance")
        img2 = gr.Image(type="filepath")
        btn2.click(plot_task_performance, outputs=img2)

    with gr.Row():
        btn3 = gr.Button("Task-Specific Top Models")
        img3 = gr.Image(type="filepath")
        btn3.click(plot_task_specific_top_models, outputs=img3)

    with gr.Row():
        btn4 = gr.Button("Top 3 Models Per Task")
        output4 = gr.Dataframe()
        btn4.click(top_3_models_per_task, outputs=output4)
        
    with gr.Row():
        btn1 = gr.Button("Show Summary Statistics")
        stats_output = gr.Dataframe()
        btn1.click(summary_statistics, outputs=stats_output)

    with gr.Row():
        btn2 = gr.Button("Plot Score Distributions")
        dist_img = gr.Image(type="filepath")
        btn2.click(plot_distribution_boxplots, outputs=dist_img)

    with gr.Row():
        btn3 = gr.Button("Best Overall Model")
        best_output = gr.Textbox()
        btn3.click(best_overall_model, outputs=best_output)

    with gr.Row():
        btn4 = gr.Button("Plot Performance Heatmap")
        heatmap_img = gr.Image(type="filepath")
        btn4.click(plot_heatmap, outputs=heatmap_img)

demo.launch()