app.py

import gradio as gr
import pandas as pd
import requests

CHUNK_SIZE = 1000
DATA_URL = "https://erkhov.com/huggingspace_data"
TIME_URL = "https://erkhov.com/huggingspace_time"

def fetch_time():
    response = requests.get(TIME_URL)
    return response.text.strip()  # assume the endpoint returns a raw timestamp or numeric value as string

def fetch_data():
    response = requests.get(DATA_URL)
    data = response.json()
    return data

def clickable(x, which_one):
    if x in ["Not Found", "Unknown"]:
        return "Not Found"
    if which_one == "models":
        return f'<a target="_blank" href="https://huggingface.co/{x}" style="color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;">{x}</a>'
    else:
        return f'<a target="_blank" href="https://huggingface.co/{which_one}/{x}" style="color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;">{x}</a>'

def create_dataframes(data):
    models_data = data["models"]
    authors_data = data["authors"]
    
    # Create DataFrame for models
    models_df = pd.DataFrame(models_data)
    models_df.rename(columns={
        "id": "Model ID",
        "author": "Author Name",
        "downloads": "Downloads (30d)",
        "likes": "Likes",
        "created_at": "Created At",
        "last_modified": "Last Modified"
    }, inplace=True)
    models_df["Model ID"] = models_df["Model ID"].apply(lambda x: clickable(x, "models"))
    models_df["Author Name"] = models_df["Author Name"].apply(lambda x: clickable(x, "models"))
    models_df = models_df.sort_values(by="Downloads (30d)", ascending=False)

    # Create DataFrame for authors
    authors_df = pd.DataFrame(authors_data)
    authors_df.rename(columns={
        "author": "Author Name",
        "models_count": "Models_Count",
        "downloads": "Total_Downloads",
        "likes": "Total_Likes"
    }, inplace=True)
    authors_df["Author Name"] = authors_df["Author Name"].apply(lambda x: clickable(x, "models"))
    authors_df = authors_df.sort_values(by="Models_Count", ascending=False)

    return models_df, authors_df

def apply_model_filters(models_df, search_query, min_downloads, min_likes):
    df = models_df.copy()
    # Extract visible text for filtering
    visible_model_id = df["Model ID"].str.extract(r'>(.*?)<')[0]
    visible_author_name = df["Author Name"].str.extract(r'>(.*?)<')[0]

    # Search filter
    if search_query.strip():
        mask = (visible_model_id.str.contains(search_query, case=False, na=False)) | \
               (visible_author_name.str.contains(search_query, case=False, na=False))
        df = df[mask]

    # Minimum downloads filter
    if min_downloads is not None and min_downloads > 0:
        df = df[df["Downloads (30d)"] >= min_downloads]

    # Minimum likes filter
    if min_likes is not None and min_likes > 0:
        df = df[df["Likes"] >= min_likes]

    return df

def filter_models(models_df, search_query, min_downloads, min_likes):
    filtered = apply_model_filters(models_df, search_query, min_downloads, min_likes)
    return filtered.iloc[:CHUNK_SIZE], CHUNK_SIZE, filtered

def update_model_table(start_idx, filtered_df):
    new_end = start_idx + CHUNK_SIZE
    combined_df = filtered_df.iloc[:new_end].copy()
    return combined_df, new_end

def apply_author_filters(authors_df, search_query, min_author_downloads, min_author_likes):
    df = authors_df.copy()

    # Extract visible text for author filtering:
    visible_author_name = df["Author Name"].str.extract(r'>(.*?)<')[0]

    # Search filter for authors
    if search_query.strip():
        mask = visible_author_name.str.contains(search_query, case=False, na=False)
        df = df[mask]

    # Minimum total downloads filter
    if min_author_downloads is not None and min_author_downloads > 0:
        df = df[df["Total_Downloads"] >= min_author_downloads]

    # Minimum total likes filter
    if min_author_likes is not None and min_author_likes > 0:
        df = df[df["Total_Likes"] >= min_author_likes]

    return df

def filter_authors(authors_df, author_search_query, min_author_downloads, min_author_likes):
    filtered_authors = apply_author_filters(authors_df, author_search_query, min_author_downloads, min_author_likes)
    return filtered_authors

def refresh_data(last_time, models_df_state, authors_df_state, stats_markdown, model_table, author_table):
    # Check if time changed
    current_time = fetch_time()
    if current_time != last_time and current_time != 0:
        # Time changed, re-fetch data
        data = fetch_data()
        models_df, authors_df = create_dataframes(data)
        
        total_models_count = data["total_models"]
        total_downloads = data["total_downloads"]
        total_likes = models_df["Likes"].sum() if "Likes" in models_df.columns else 0
        
        # Update stats markdown
        new_stats_markdown = f"""
        # GGUF Models and Authors Leaderboard
        **Total Models:** {total_models_count} | **Total Downloads (30d):** {total_downloads} | **Total Likes:** {total_likes}  
        **Last Updated:** {current_time}
        """
        
        # Update states
        return current_time, models_df, authors_df, gr.update(value=new_stats_markdown), gr.update(value=models_df.iloc[:CHUNK_SIZE]), gr.update(value=authors_df)
    else:
        # No change
        return last_time, models_df_state, authors_df_state, stats_markdown, model_table, author_table

# Initial fetch
initial_time = fetch_time()
data = fetch_data()
all_models_df, authors_df = create_dataframes(data)

total_models_count = data["total_models"]
total_downloads = data["total_downloads"]
total_likes = all_models_df["Likes"].sum() if "Likes" in all_models_df.columns else 0

initial_stats_markdown = f"""
# GGUF Models and Authors Leaderboard
**Total Models:** {total_models_count} | **Total Downloads (30d):** {total_downloads} | **Total Likes:** {total_likes}  
**Last Updated:** {initial_time}
"""

with gr.Blocks() as demo:
    gr.Markdown(f"""
    # 🚀GGUF Tracker🚀
    Welcome to 🚀**GGUF Tracker**🚀, a live-updating leaderboard for all things GGUF on 🚀Hugging Face.  
    Stats refresh every hour on the backend, but this interface checks every 5 minutes for updates.

    By the way, I’m 🚀Richard Erkhov, and you can check out more of what I’m working on at my [🌟**github**](https://github.com/RichardErkhov), 
    [🌟**huggingface**](https://huggingface.co/RichardErkhov) or [🌟**erkhov.com**](https://erkhov.com). Go take a look—I think you’ll like what you find.
    """)

    stats_markdown = gr.Markdown(initial_stats_markdown)
    
    with gr.Tabs():
        with gr.TabItem("Models"):
            with gr.Row():
                search_query = gr.Textbox(label="Search (by Model ID or Author Name)")
                min_downloads = gr.Number(label="Min Downloads (30d)", value=0)
                min_likes = gr.Number(label="Min Likes", value=0)

            filter_button = gr.Button("Apply Filters")
            model_table = gr.DataFrame(
                value=all_models_df.iloc[:CHUNK_SIZE],
                interactive=False,
                label="GGUF Models (Click column headers to sort)",
                wrap=True,
                datatype=["markdown", "markdown", "number", "number", "str", "str"]
            )
            load_more_button = gr.Button("Load More Models")
            
            # States for models
            start_idx = gr.State(value=CHUNK_SIZE)
            filtered_df_state = gr.State(value=all_models_df)

            filter_button.click(
                fn=filter_models,
                inputs=[filtered_df_state, search_query, min_downloads, min_likes],
                outputs=[model_table, start_idx, filtered_df_state]
            )
            load_more_button.click(fn=update_model_table, inputs=[start_idx, filtered_df_state], outputs=[model_table, start_idx])

        with gr.TabItem("Authors"):
            with gr.Row():
                author_search_query = gr.Textbox(label="Search by Author Name")
                min_author_downloads = gr.Number(label="Min Total Downloads", value=0)
                min_author_likes = gr.Number(label="Min Total Likes", value=0)

            author_filter_button = gr.Button("Apply Filters")
            author_table = gr.DataFrame(
                value=authors_df,
                interactive=False,
                label="Authors (Click column headers to sort)",
                wrap=True,
                datatype=["markdown", "number", "number", "number"]
            )

            author_filter_button.click(
                fn=filter_authors,
                inputs=[authors_df, author_search_query, min_author_downloads, min_author_likes],
                outputs=author_table
            )

    # States for refresh
    last_hf_time_state = gr.State(value=initial_time)
    models_df_state = gr.State(value=all_models_df)
    authors_df_state = gr.State(value=authors_df)

    # Timer to check every 5 minutes = 300 seconds
    timer = gr.Timer(interval=300, fn=refresh_data, inputs=[
        last_hf_time_state,
        models_df_state,
        authors_df_state,
        stats_markdown,
        model_table,
        author_table
    ], outputs=[
        last_hf_time_state,
        models_df_state,
        authors_df_state,
        stats_markdown,
        model_table,
        author_table
    ])

    demo.launch()