import gradio as gr
import requests
from bs4 import BeautifulSoup
from transformers import pipeline

task_extractor = pipeline("text2text-generation", model="google/flan-t5-small")

# Simulated LLM task extraction (replace with real call if local)
def extract_task(user_input):
    prompt = f"Classify the following ML task: {user_input}. Just reply with the task name."
    result = task_extractor(prompt, max_new_tokens=10)
    task = result[0]["generated_text"].strip().lower()
    return task


# Scrape Hugging Face models by task
def get_models_for_task(task):
    url = f"https://huggingface.co/models?pipeline_tag={task}"
    headers = {"User-Agent": "Mozilla/5.0"}
    try:
        res = requests.get(url, headers=headers)
        soup = BeautifulSoup(res.text, "html.parser")

        results = []
        for a in soup.find_all("a", class_="flex items-center gap-2"):
            model_name = a.get("href", "").strip("/").split("/")[-1]
            if model_name:
                results.append(model_name)
            if len(results) >= 5:
                break

        return results if results else ["No models found"]

    except Exception as e:
        return [f"Error: {str(e)}"]

# Combined predict function
def predict(user_input):
    task = extract_task(user_input)
    models = get_models_for_task(task)
    return f"🧠 Task: {task}\n📦 Models:\n" + "\n".join(models)

# Gradio UI
with gr.Blocks() as demo:
    gr.Markdown("## 🤖 ML Task → Hugging Face Model Finder")
    with gr.Row():
        input_box = gr.Textbox(label="Describe your ML task")
        submit_btn = gr.Button("🔍 Find Models")
    output_box = gr.Textbox(label="Suggested Models", lines=10)
    submit_btn.click(predict, inputs=input_box, outputs=output_box)

demo.launch()