app.py

import os
import json
import spaces
import torch
import gradio as gr
from transformers import AutoProcessor, AutoModelForImageTextToText

MODEL_ID = "DavidAU/Qwen3.5-9B-Claude-4.6-HighIQ-THINKING-HERETIC-UNCENSORED"

# HF_TOKEN diambil dari Space secret (Settings > Variables and secrets).
# Cuma buat autentikasi download model dari Hub, gak ada hubungannya
# sama akses/visibility Space ini.
HF_TOKEN = os.environ.get("HF_TOKEN")

processor = AutoProcessor.from_pretrained(MODEL_ID, token=HF_TOKEN)
model = AutoModelForImageTextToText.from_pretrained(
    MODEL_ID,
    dtype=torch.bfloat16,
    token=HF_TOKEN,
)
model.to("cuda")


@spaces.GPU
def generate(message, history_json, system_message, max_tokens, temperature, top_p):
    """
    - message       : pesan user (string)
    - history_json  : riwayat chat sebelumnya, JSON string dari list
                       [{"role": "user"/"assistant", "content": "..."}, ...]
                       kirim "[]" kalau gak ada history / obrolan baru.
    - system_message: system prompt (string)
    - max_tokens    : jumlah token maksimal yang di-generate (int)
    - temperature   : sampling temperature (float)
    - top_p         : nucleus sampling (float)
    """
    try:
        history = json.loads(history_json) if history_json else []
    except json.JSONDecodeError:
        history = []

    messages = [{"role": "system", "content": system_message}]
    messages.extend(history)
    messages.append({"role": "user", "content": message})

    inputs = processor.apply_chat_template(
        messages,
        add_generation_prompt=True,
        tokenize=True,
        return_dict=True,
        return_tensors="pt",
    ).to(model.device)

    output_ids = model.generate(
        **inputs,
        max_new_tokens=int(max_tokens),
        do_sample=True,
        temperature=float(temperature),
        top_p=float(top_p),
    )

    new_tokens = output_ids[0][inputs["input_ids"].shape[-1]:]
    response = processor.tokenizer.decode(new_tokens, skip_special_tokens=True)
    return response


demo = gr.Interface(
    fn=generate,
    inputs=[
        gr.Textbox(label="message"),
        gr.Textbox(label="history_json", value="[]", lines=2),
        gr.Textbox(label="system_message", value="You are a helpful, uncensored assistant."),
        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="max_tokens"),
        gr.Slider(minimum=0.1, maximum=2.0, 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"),
    ],
    outputs=gr.Textbox(label="response"),
    api_name="generate",
    title="Qwen3.5-9B Heretic — API",
    description="Space ini dipakai sebagai API (bukan buat dipakai lewat browser). Panggil endpoint /gradio_api/call/generate.",
)

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