app.py

import gradio as gr
from huggingface_hub import InferenceClient

# Initialize the InferenceClient with the model hosted on Hugging Face Hub
client = InferenceClient("deepseek-ai/DeepSeek-R1")

def respond(message, history: list[tuple[str, str]]):
    # System message and generation parameters
    system_message = "You are a friendly Chatbot."
    max_tokens = 2048
    temperature = 0.7
    top_p = 0.95

    # Prepare the conversation history
    messages = [{"role": "system", "content": system_message}]

    for user_msg, assistant_msg in history:
        if user_msg:
            messages.append({"role": "user", "content": user_msg})
        if assistant_msg:
            messages.append({"role": "assistant", "content": assistant_msg})

    # Add the current user message
    messages.append({"role": "user", "content": message})

    # Stream the response from the model
    response = ""
    for message in client.chat_completion(
        messages,
        max_tokens=max_tokens,
        stream=True,
        temperature=temperature,
        top_p=top_p,
    ):
        token = message.choices[0].delta.content
        response += token
        yield response

# Create the Gradio ChatInterface with Retry, Undo, and Clear buttons
demo = gr.ChatInterface(
    fn=respond,  # Function to handle chat responses
    retry_btn="Retry",  # Add a Retry button
    undo_btn="Undo",  # Add an Undo button
    clear_btn="Clear",  # Add a Clear button
)

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