Image:
  from_base: parachutes/base-python:3.12.9
  run_command:
    - pip install --no-cache-dir torch==2.10.0 torchaudio==2.10.0 --extra-index-url https://download.pytorch.org/whl/cu128
    - pip install --no-cache-dir transformers==4.57.3 accelerate==1.12.0
    - pip install --no-cache-dir qwen-tts==0.1.1
    - pip install --no-cache-dir pyloudnorm scipy numpy soundfile librosa pyyaml huggingface_hub einops
    - pip install --no-cache-dir faster-whisper jiwer
    - pip install "https://github.com/lesj0610/flash-attention/releases/download/v2.8.3-cu12-torch2.10-cp312/flash_attn-2.8.3%2Bcu12torch2.10cxx11abiTRUE-cp312-cp312-linux_x86_64.whl"
    # ── UTMOSv2 weights MUST land where chutes user will look at runtime ──
    - pip install git+https://github.com/sarulab-speech/UTMOSv2.git
    - mkdir -p /home/chutes/.cache && chown -R chutes:chutes /home/chutes 2>/dev/null || true
    - HOME=/home/chutes python -c "import utmosv2; m = utmosv2.create_model(pretrained=True); print('UTMOSv2 cached at /home/chutes/.cache/utmosv2')"
  
NodeSelector:
  gpu_count: 2
  min_vram_gb_per_gpu: 80
  include: ["pro_6000"]
  exclude: []

Chute:
  tagline: Vocence TTS — Qwen3 PromptTTS (weights in repo)
  readme: Qwen3 12Hz TTS snapshot + miner.py for Vocence
  shutdown_after_seconds: 86400
  concurrency: 2
  max_instances: 1
  scaling_threshold: 0.5
  tee: true