import spaces
import gradio as gr
from transformers import AutoModelForCausalLM, AutoTokenizer

# Load the model and tokenizer locally
model_name = "kz919/QwQ-0.5B-Distilled-SFT"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name).to("cuda")

# Define the function to handle chat responses
@spaces.GPU
def respond(message, history: list[tuple[str, str]], system_message, max_tokens, temperature, top_p):
    # Prepare the prompt by combining history and system messages
    prompt = system_message + "\n"
    for user_input, assistant_response in history:
        prompt += f"User: {user_input}\nAssistant: {assistant_response}\n"
    prompt += f"User: {message}\nAssistant:"

    # Tokenize the input prompt
    inputs = tokenizer(prompt, return_tensors="pt").to("cuda")

    # Generate a response
    outputs = model.generate(
        inputs.input_ids,
        max_length=max_tokens,
        temperature=temperature,
        top_p=top_p,
        pad_token_id=tokenizer.eos_token_id,
    )

    # Decode the generated tokens and yield the response
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
    yield response.split("Assistant:")[-1].strip()


# Create the Gradio interface
demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Textbox(value="You are a helpful and harmless assistant. You are Qwen developed by Alibaba. You should think step-by-step.", label="System message"),
        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
        gr.Slider(minimum=0.1, maximum=4.0, 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)"),
    ],
)

# Launch the Gradio app
if __name__ == "__main__":
    demo.launch()