# Comprehensive Model Performance Analysis # Importing Required Libraries import pandas as pd import matplotlib.pyplot as plt import seaborn as sns import gradio as gr import requests from bs4 import BeautifulSoup import io import os import base64 import zipfile from PIL import Image from io import BytesIO # Input Data # Input data with links to Hugging Face repositories data_full = [ ['CultriX/Qwen2.5-14B-SLERPv7', 'https://huggingface.co./CultriX/Qwen2.5-14B-SLERPv7', 0.7205, 0.8272, 0.7541, 0.6581, 0.5, 0.729], ['djuna/Q2.5-Veltha-14B-0.5', 'https://huggingface.co./djuna/Q2.5-Veltha-14B-0.5', 0.7492, 0.8386, 0.7305, 0.598, 0.43, 0.7817], ['CultriX/Qwen2.5-14B-FinalMerge', 'https://huggingface.co./CultriX/Qwen2.5-14B-FinalMerge', 0.7248, 0.8277, 0.7113, 0.7052, 0.57, 0.7001], ['CultriX/Qwen2.5-14B-MultiCultyv2', 'https://huggingface.co./CultriX/Qwen2.5-14B-MultiCultyv2', 0.7295, 0.8359, 0.7363, 0.5767, 0.44, 0.7316], ['CultriX/Qwen2.5-14B-Brocav7', 'https://huggingface.co./CultriX/Qwen2.5-14B-Brocav7', 0.7445, 0.8353, 0.7508, 0.6292, 0.46, 0.7629], ['CultriX/Qwen2.5-14B-Broca', 'https://huggingface.co./CultriX/Qwen2.5-14B-Broca', 0.7456, 0.8352, 0.748, 0.6034, 0.44, 0.7716], ['CultriX/Qwen2.5-14B-Brocav3', 'https://huggingface.co./CultriX/Qwen2.5-14B-Brocav3', 0.7395, 0.8388, 0.7393, 0.6405, 0.47, 0.7659], ['CultriX/Qwen2.5-14B-Brocav4', 'https://huggingface.co./CultriX/Qwen2.5-14B-Brocav4', 0.7432, 0.8377, 0.7444, 0.6277, 0.48, 0.758], ['CultriX/Qwen2.5-14B-Brocav2', 'https://huggingface.co./CultriX/Qwen2.5-14B-Brocav2', 0.7492, 0.8302, 0.7508, 0.6377, 0.51, 0.7478], ['CultriX/Qwen2.5-14B-Brocav5', 'https://huggingface.co./CultriX/Qwen2.5-14B-Brocav5', 0.7445, 0.8313, 0.7547, 0.6376, 0.5, 0.7304], ['CultriX/Qwen2.5-14B-Brocav6', 'https://huggingface.co./CultriX/Qwen2.5-14B-Brocav6', 0.7179, 0.8354, 0.7531, 0.6378, 0.49, 0.7524], ['CultriX/Qwenfinity-2.5-14B', 'https://huggingface.co./CultriX/Qwenfinity-2.5-14B', 0.7347, 0.8254, 0.7279, 0.7267, 0.56, 0.697], ['CultriX/Qwen2.5-14B-Emergedv2', 'https://huggingface.co./CultriX/Qwen2.5-14B-Emergedv2', 0.7137, 0.8335, 0.7363, 0.5836, 0.44, 0.7344], ['CultriX/Qwen2.5-14B-Unity', 'https://huggingface.co./CultriX/Qwen2.5-14B-Unity', 0.7063, 0.8343, 0.7423, 0.682, 0.57, 0.7498], ['CultriX/Qwen2.5-14B-MultiCultyv3', 'https://huggingface.co./CultriX/Qwen2.5-14B-MultiCultyv3', 0.7132, 0.8216, 0.7395, 0.6792, 0.55, 0.712], ['CultriX/Qwen2.5-14B-Emergedv3', 'https://huggingface.co./CultriX/Qwen2.5-14B-Emergedv3', 0.7436, 0.8312, 0.7519, 0.6585, 0.55, 0.7068], ['CultriX/SeQwence-14Bv1', 'https://huggingface.co./CultriX/SeQwence-14Bv1', 0.7278, 0.841, 0.7541, 0.6816, 0.52, 0.7539], ['CultriX/Qwen2.5-14B-Wernickev2', 'https://huggingface.co./CultriX/Qwen2.5-14B-Wernickev2', 0.7391, 0.8168, 0.7273, 0.622, 0.45, 0.7572], ['CultriX/Qwen2.5-14B-Wernickev3', 'https://huggingface.co./CultriX/Qwen2.5-14B-Wernickev3', 0.7357, 0.8148, 0.7245, 0.7023, 0.55, 0.7869], ['CultriX/Qwen2.5-14B-Wernickev4', 'https://huggingface.co./CultriX/Qwen2.5-14B-Wernickev4', 0.7355, 0.829, 0.7497, 0.6306, 0.48, 0.7635], ['CultriX/SeQwential-14B-v1', 'https://huggingface.co./CultriX/SeQwential-14B-v1', 0.7355, 0.8205, 0.7549, 0.6367, 0.48, 0.7626], ['CultriX/Qwen2.5-14B-Wernickev5', 'https://huggingface.co./CultriX/Qwen2.5-14B-Wernickev5', 0.7224, 0.8272, 0.7541, 0.679, 0.51, 0.7578], ['CultriX/Qwen2.5-14B-Wernickev6', 'https://huggingface.co./CultriX/Qwen2.5-14B-Wernickev6', 0.6994, 0.7549, 0.5816, 0.6991, 0.58, 0.7267], ['CultriX/Qwen2.5-14B-Wernickev7', 'https://huggingface.co./CultriX/Qwen2.5-14B-Wernickev7', 0.7147, 0.7599, 0.6097, 0.7056, 0.57, 0.7164], ['CultriX/Qwen2.5-14B-FinalMerge-tmp2', 'https://huggingface.co./CultriX/Qwen2.5-14B-FinalMerge-tmp2', 0.7255, 0.8192, 0.7535, 0.6671, 0.5, 0.7612], ] columns = ["Model Configuration", "Model Link", "tinyArc", "tinyHellaswag", "tinyMMLU", "tinyTruthfulQA", "tinyTruthfulQA_mc1", "tinyWinogrande"] df_full = pd.DataFrame(data_full, columns=columns) # Visualization and Analytics Functions # 1. Plot Average Scores def plot_average_scores(): df_full["Average Score"] = df_full.iloc[:, 2:].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.show() # 2. Plot Task Performance def plot_task_performance(): df_full_melted = df_full.melt(id_vars=["Model Configuration", "Model Link"], 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.show() # 3. Plot Task-Specific Top Models def plot_task_specific_top_models(): top_models = df_full.iloc[:, 2:].idxmax() top_scores = df_full.iloc[:, 2:].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.show() # YAML Configuration and Scraping Utilities # 1. Scrape MergeKit Configuration def scrape_mergekit_config(model_name): model_link = df_full.loc[df_full["Model Configuration"] == model_name, "Model Link"].values[0] response = requests.get(model_link) if response.status_code != 200: return f"Failed to fetch model page for {model_name}. Please check the link." soup = BeautifulSoup(response.text, "html.parser") yaml_config = soup.find("pre") return yaml_config.text.strip() if yaml_config else f"No YAML configuration found for {model_name}." # 2. Download All Data def download_all_data(): csv_buffer = io.StringIO() df_full.to_csv(csv_buffer, index=False) csv_data = csv_buffer.getvalue().encode('utf-8') zip_buffer = io.BytesIO() with zipfile.ZipFile(zip_buffer, 'w') as zf: zf.writestr("model_scores.csv", csv_data) zip_buffer.seek(0) return zip_buffer, "analysis_data.zip" # Performance Heatmap def plot_heatmap(): plt.figure(figsize=(12, 8)) sns.heatmap(df_full.iloc[:, 2:], annot=True, cmap="YlGnBu", xticklabels=columns[2:], yticklabels=df_full["Model Configuration"]) plt.title("Performance Heatmap", fontsize=16) plt.tight_layout() plt.show() # Gradio App # Building the Interface with gr.Blocks() as demo: gr.Markdown("# Comprehensive Model Performance Analysis with Hugging Face Links") with gr.Row(): btn1 = gr.Button("Show Average Performance") img1 = gr.Image(type="pil", label="Average Performance Plot") btn1.click(plot_average_scores, outputs=[img1]) with gr.Row(): btn2 = gr.Button("Show Task Performance") img2 = gr.Image(type="pil", label="Task Performance Plot") btn2.click(plot_task_performance, outputs=[img2]) with gr.Row(): btn3 = gr.Button("Task-Specific Top Models") img3 = gr.Image(type="pil", label="Task-Specific Top Models Plot") btn3.click(plot_task_specific_top_models, outputs=[img3]) with gr.Row(): btn4 = gr.Button("Plot Performance Heatmap") img4 = gr.Image(type="pil", label="Performance Heatmap") btn4.click(plot_heatmap, outputs=[img4]) with gr.Row(): download_all_btn = gr.Button("Download Everything") all_downloads = gr.File(label="Download All Data") download_all_btn.click(download_all_data, outputs=all_downloads) demo.launch()