app.py

import gradio as gr
from transformers import pipeline
import torch
import os

# Configure cache to avoid space limitations
os.environ['HF_HOME'] = '/tmp/cache'

# Use a reliable LLM hosted by Hugging Face
MODEL_NAME = "mistralai/Mistral-7B-Instruct-v0.2"

# Load the model pipeline
generator = pipeline(
    "text-generation",
    model=MODEL_NAME,
    device_map="auto",
    torch_dtype=torch.bfloat16,
    max_new_tokens=560
)

def generate_chat_completion(message_history, max_tokens=560, temperature=0.8):
    """Generate assistant response from chat message history"""
    try:
        # If using Gradio chat format (list of tuples), convert to role-content dicts
        messages = [{"role": "user", "content": msg} if i % 2 == 0 else {"role": "assistant", "content": msg} 
                    for i, msg in enumerate(message_history)]
        
        prompt = "\n".join([f"{m['role'].capitalize()}: {m['content']}" for m in messages])
        prompt += "\nAssistant:"
        
        output = generator(
            prompt,
            max_new_tokens=max_tokens,
            temperature=temperature,
            top_p=0.95,
            repetition_penalty=1.15,
            do_sample=True
        )
        response = output[0]['generated_text'].replace(prompt, "").strip()
        return message_history + [response]
    except Exception as e:
        return message_history + [f"[Error] {str(e)}"]

# Gradio Chat Interface
chat_interface = gr.ChatInterface(
    fn=generate_chat_completion,
    title="Mistral-7B Chat",
    description="Powered by Hugging Face Transformers",
    retry_btn="Retry",
    undo_btn="Undo",
    clear_btn="Clear"
)

if __name__ == "__main__":
    chat_interface.launch()