app.py

import os
import gradio as gr
from huggingface_hub import InferenceClient

MODEL_ID = "unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF"

# Pega o token do secret HF_TOKEN que você adicionou no Space
token = os.environ.get("HF_TOKEN")

# Inicializa o cliente; se token for None, InferenceClient tentará usar o token local/config.
client = InferenceClient(model=MODEL_ID, token=token)

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

    # MODE: escolha "stream_mode = True" para token por token, ou False para resposta completa de uma vez
    stream_mode = True

    if stream_mode:
        response = ""
        # stream=True entrega chunks — iteramos e extraímos 'content' do delta
        for chunk in client.chat_completion(
            messages,
            max_tokens=max_tokens,
            stream=True,
            temperature=temperature,
            top_p=top_p,
        ):
            # chunk pode ser dataclass/obj ou dict-like; tentamos extrair o texto com segurança
            token_piece = ""
            try:
                delta = chunk.choices[0].delta
                if isinstance(delta, dict):
                    token_piece = delta.get("content", "") or ""
                else:
                    # objeto dataclass-like
                    token_piece = getattr(delta, "content", "") or ""
            except Exception:
                # fallback genérico (caso a API retorne formato diferente)
                token_piece = str(chunk)

            response += token_piece
            yield response

    else:
        # Sem streaming: recupera a resposta completa
        completion = client.chat_completion(
            messages,
            max_tokens=max_tokens,
            stream=False,
            temperature=temperature,
            top_p=top_p,
        )
        # conforme docs, a resposta completa aparece em:
        text = completion.choices[0].message.content
        yield text


demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Textbox(value="You are a friendly Chatbot.", 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)"),
    ],
)

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