app.py

import gradio as gr
from huggingface_hub import InferenceClient

# Use the CRA-v1-7B model (which uses the GGUF file internally)
client = InferenceClient("molbal/CRA-v1-7B")

def respond(message, history: list[tuple[str, str]], system_message, max_tokens, temperature, top_p):
    # Build the conversation history; always include the system message
    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})
    messages.append({"role": "user", "content": message})
    
    response = ""
    # Call the model with streaming and the new parameters
    for chunk in client.chat_completion(
        messages,
        max_tokens=max_tokens,
        stream=True,
        temperature=temperature,
        top_p=top_p,
        num_ctx=16384,
        repeat_penalty=1.05,
    ):
        token = chunk.choices[0].delta.content
        response += token
        yield response

# Create an alert message to inform users that inference runs on CPU (and will be slow)
cpu_alert = gr.Markdown("**Note:** This model runs on CPU, so inference may be slow.")

# Build the UI using Blocks to combine the alert and the ChatInterface
with gr.Blocks() as demo:
    cpu_alert.render()
    chat_interface = gr.ChatInterface(
        respond,
        additional_inputs=[
            gr.Textbox(
                value="### System: You are a writer’s assistant. ### Task: Understand how the story flows, what motivations the characters have and how they will interact with each other and the world as a step by step thought process before continuing the story. ### Context: {context}",
                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=0.8, step=0.05, label="Top-p (nucleus sampling)")
        ]
    )
    chat_interface.render()

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