from huggingface_hub import HfApi, HfFileSystem
import re
from tqdm import tqdm
import concurrent.futures
import gradio as gr
import datetime
import pandas as pd
import os
import threading
import time

HF_TOKEN = os.getenv('HF_TOKEN')

api = HfApi()
fs = HfFileSystem()

def restart_space():
    time.sleep(36000)
    api.restart_space(repo_id="Tanvir1337/mradermacher-quantized-models", token=HF_TOKEN)

text = f"""
🎯 The Leaderboard aims to track mradermacher's gguf quantized models.

## 🛠️ Backend

The leaderboard's backend mainly runs on the [Hugging Face Hub API](https://huggingface.co/docs/huggingface_hub/v0.5.1/en/package_reference/hf_api).

## 🔍 Searching

You can search for author or a spesific model using the search bar.

## ⌛ Last Update

This space is last updated in **{str(datetime.datetime.now().strftime("%Y-%m-%d %H:%M"))}**.

## 📒 Important Note

This space potentially includes incorrectly quantized models for a model.

If you find any incorrectly quantized model, please report it to me.
"""

quant_models = [i.__dict__['id'] for i in api.list_models(author="mradermacher") if "GGUF" in i.__dict__['id']]

pattern = r'\(https://huggingface\.co/([^/]+)/([^/]+)\)'
liste = {}

def process_model(i, pattern, liste):
    text = fs.read_text(i + "/README.md")
    matches = re.search(pattern, text)

    if matches:
        author = matches.group(1)
        model_name = matches.group(2)
        full_id = (author + "/" + model_name).split(")")[0]

        try:
            liste[full_id].append(i)
        except KeyError:
            liste[full_id] = [i]


num_threads = 64

with concurrent.futures.ThreadPoolExecutor(max_workers=num_threads) as executor:
    futures = []
    for i in quant_models:
        future = executor.submit(process_model, i, pattern, liste)
        futures.append(future)

    concurrent.futures.wait(futures)


authors, models, gguf = [], [], []


for model, values in liste.items():
    models.append(model)

    gguf_value = None

    for value in values:
        if "-GGUF" in value:
            gguf_value = value

    authors.append(model.split('/')[0])
    gguf.append(gguf_value)


df = pd.DataFrame({'👤 Author Name': authors, '🤖 Model Name': models, '📥 GGUF': gguf})


def search(search_text):
    if not search_text:
        return df

    if len(search_text.split('/'))>1:
        return df[df['🤖 Model Name'] == clickable(search_text)]
    else:
        return df[df['👤 Author Name'] == clickable(search_text)]


def clickable(x):
    return None if not x else f'<a target="_blank" href="https://huggingface.co/{x}" style="color: var(--link-text-color); text-decoration: underline;text-decoration-style: dotted;">{x}</a>'


def to_clickable(df):
    for column in list(df.columns):
        df[column] = df[column].apply(lambda x: clickable(x))
    return df


with gr.Blocks() as demo:
    gr.Markdown("""<center><img src = "https://huggingface.co/avatars/6b97d30ff0bdb5d5c633ba850af739cd.svg" width=200 height=200></center>""")
    gr.Markdown("""<h1 align="center" id="space-title">mradermacher Quantized Models</h1>""")
    gr.Markdown(text)

    with gr.Column(min_width=320):
        search_bar = gr.Textbox(placeholder="🔍 Search for a author or a specific model", show_label=False)

    
    df_clickable = to_clickable(df)
    gr_df = gr.Dataframe(df_clickable, interactive=False, datatype=["markdown"]*len(df.columns))

    search_bar.submit(fn=search, inputs=search_bar, outputs=gr_df)

threading.Thread(target=restart_space).start()
demo.launch()