app.py

"""RAP RL checkpoint-240 on ZeroGPU; BF16 weights, no quantization."""
import spaces
import os
os.environ.setdefault('HF_HUB_DISABLE_PROGRESS_BARS','1')
os.environ.setdefault('TOKENIZERS_PARALLELISM','false')
import json
import shutil
import time
from pathlib import Path
import torch
import gradio as gr
import psutil
from transformers import AutoModelForImageTextToText, AutoProcessor
from demo_ui import build

REPO = "VincentHancoder/RAP-Qwen3.8-27B-RL"
REVISION = "50f3adc4ebb9b19e53ef9c21c293993a77bf3fd7"
print("STARTUP", json.dumps({"ram_gib": psutil.virtual_memory().total / 2**30,
                            "disk_free_gib": shutil.disk_usage(".").free / 2**30,
                            "torch": torch.__version__}), flush=True)
processor = AutoProcessor.from_pretrained(REPO, revision=REVISION, token=os.environ.get("HF_TOKEN"))
model = AutoModelForImageTextToText.from_pretrained(
    REPO, revision=REVISION, token=os.environ.get("HF_TOKEN"), dtype=torch.bfloat16,
    device_map="cuda:0", attn_implementation="sdpa",
)
model.eval()
model.config.use_cache = True
model.config.text_config.use_cache = True
print("RAP_MODEL_READY", flush=True)

demo, artifacts = build(model, processor)
demo.queue(default_concurrency_limit=1, max_size=8).launch(
    show_error=False, max_file_size='40mb', ssr_mode=False,
    allowed_paths=[str(artifacts),str(Path(__file__).parent/'examples')],
    blocked_paths=[str(__import__('pathlib').Path(__file__).parent/'.runtime'),
                   '/root/.cache/huggingface','/home/user/.cache/huggingface',
                   '/etc','/proc','/sys'],
)