app.py

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

# 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
    msg = [
        {"role": "system", "content": system_message}
    ]
    for user_input, assistant_response in history:
        msg.extend(
            [
                {"role": "user", "content": user_input},
                {"role": "assistant", "content": assistant_response}
            ]
        )
    msg.append({"role": "user", "content": message})

    prompt = tokenizer.apply_chat_template(
        msg,
        tokenize=False,
        add_generation_prompt=True
    )

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


    streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)

    # Use a thread to run the generation in parallel
    generation_thread = threading.Thread(
        target=model.generate,
        kwargs=dict(
            inputs=inputs.input_ids,
            max_length=max_tokens,
            streamer=streamer,
            do_sample=True,
            temperature=temperature,
            top_p=top_p,
            pad_token_id=tokenizer.eos_token_id,
        ),
    )
    generation_thread.start()

    # Stream the tokens as they are generated
    text_buffer = ""
    for new_text in streamer:
        text_buffer+=new_text
        yield text_buffer


# 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=16384, 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()