app.py

import gradio as gr
import os
from threading import Thread
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
from peft import PeftModel

# --- 1. Load your Fine-Tuned Model and Tokenizer ---
# Make sure to set your HUGGINGFACEHUB_API_TOKEN in your Space's secrets
api_token = os.getenv("HUGGINGFACEHUB_API_TOKEN")

if not api_token:
    raise ValueError("❌ ERROR: Hugging Face API token is not set. Please set it in your Space secrets.")

# Define model names
base_model_name = "unsloth/qwen2.5-math-7b-bnb-4bit"
peft_model_name = "Hrushi02/Root_Math"

# Load base model and tokenizer
base_model = AutoModelForCausalLM.from_pretrained(
    base_model_name,
    torch_dtype=torch.float16,
    device_map="auto",
    token=api_token
)
tokenizer = AutoTokenizer.from_pretrained(base_model_name, token=api_token)

# Load your fine-tuned PEFT model
model = PeftModel.from_pretrained(base_model, peft_model_name, token=api_token)
print("✅ Model loaded successfully!")

# --- 2. Rewrite the Respond Function to Use YOUR Model ---
def respond(
    message,
    history: list[tuple[str, str]],
    system_message,
    max_tokens,
    temperature,
    top_p,
):
    # Create the chat history format
    messages = [{"role": "system", "content": system_message}]
    for val in history:
        if val[0]:
            messages.append({"role": "user", "content": val[0]})
        if val[1]:
            messages.append({"role": "assistant", "content": val[1]})
    messages.append({"role": "user", "content": message})

    # Prepare for streaming
    streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
    
    # Tokenize the input
    inputs = tokenizer.apply_chat_template(
        messages,
        add_generation_prompt=True,
        return_tensors="pt"
    ).to(model.device)

    # Generation arguments
    generation_kwargs = dict(
        inputs=inputs,
        streamer=streamer,
        max_new_tokens=max_tokens,
        temperature=temperature,
        top_p=top_p,
        do_sample=True,
    )

    # Start generation in a separate thread
    thread = Thread(target=model.generate, kwargs=generation_kwargs)
    thread.start()

    # Yield the generated tokens
    response = ""
    for token in streamer:
        response += token
        yield response

# --- 3. Launch the Gradio Interface (No Changes Here) ---
demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Textbox(value="You are a math assistant. Solve the following math problem.", label="System message"),
        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
        gr.Slider(minimum=0.1, maximum=1.0, value=0.7, step=0.1, label="Temperature"),
        gr.Slider(
            minimum=0.1,
            maximum=1.0,
            value=0.95,
            step=0.05,
            label="Top-p (nucleus sampling)",
        ),
    ],
)

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