app.py

import re
import gradio as gr
from llama_cpp import Llama

model = "ggml-org/gemma-3-1b-it-GGUF"
llm = Llama.from_pretrained(
    repo_id=model,
    filename="gemma-3-1b-it-Q8_0.gguf",
    verbose=True,
    use_mmap=True,
    use_mlock=True,
    n_threads=4,
    n_threads_batch=4,
    n_ctx=8000,
)


def respond(
    message,
    history: list[tuple[str, str]],
    system_message,
    max_tokens,
    temperature,
    top_p,
):
    if len(system_message) > 0:
        messages = [{"role": "system", "content": system_message}]
    else:
        messages = []
        

    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})

    response = ""
    completion = llm.create_chat_completion(
        messages,
        max_tokens=max_tokens,
        stream=True,
        temperature=temperature,
        top_p=top_p
    )

    for message in completion:
        delta = message['choices'][0]['delta']
        if 'content' in delta:
            response += delta['content']
            formatted_response = re.sub(r"<think>\s*(.*?)\s*</think>", r"*\1*", response, flags=re.DOTALL)
            yield formatted_response


demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Textbox(
            value="",
            label="System message",
        ),
        gr.Slider(minimum=200, maximum=100000, value=4000, step=100, label="Max new tokens"),
        gr.Slider(minimum=0.1, maximum=4.0, value=0.6, step=0.1, label="Temperature"),
        gr.Slider(
            minimum=0.1,
            maximum=1.0,
            value=0.95,
            step=0.05,
            label="Top-p (nucleus sampling)",
        ),
    ],
    description=model,
)


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