app.py

# app.py
import os
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
import gradio as gr

MODEL_ID = "mistralai/Mistral-7B-Instruct-v0.1"

# Optional: get token from environment
HF_TOKEN = os.getenv("HF_TOKEN")  # make sure you added this secret in HF Space

# Load model & tokenizer
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, token=HF_TOKEN)
model = AutoModelForCausalLM.from_pretrained(
    MODEL_ID,
    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
    device_map="auto",
    token=HF_TOKEN
)

def chat_with_mistral(prompt, history=None):
    history = history or []
    # Build conversation prompt manually
    full_prompt = ""
    for user_text, bot_text in history:
        full_prompt += f"User: {user_text}\nAssistant: {bot_text}\n"
    full_prompt += f"User: {prompt}\nAssistant:"

    inputs = tokenizer(full_prompt, return_tensors="pt").to(model.device)
    with torch.no_grad():
        outputs = model.generate(
            **inputs,
            max_new_tokens=200,
            do_sample=True,
            top_p=0.95,
            temperature=0.7,
            pad_token_id=tokenizer.eos_token_id
        )
    decoded = tokenizer.decode(outputs[0], skip_special_tokens=True)
    
    # Extract only the assistant reply
    answer = decoded.split("Assistant:")[-1].strip()
    history.append((prompt, answer))
    return answer, history


with gr.Blocks(title="Mistral Chatbot") as app:
    gr.Markdown("# 🐦 Mistral Chatbot (with HF Token)")
    chatbot = gr.Chatbot()
    msg = gr.Textbox(label="Your message")
    clear = gr.Button("Clear")
    
    conversation_state = gr.State([])

    def handle_user_message(user_message, history):
        answer, new_history = chat_with_mistral(user_message, history)
        return "", new_history
    
    msg.submit(handle_user_message, [msg, conversation_state], [msg, chatbot, conversation_state])
    clear.click(lambda: ([], []), inputs=None, outputs=[chatbot, conversation_state])

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