| Image: | |
| from_base: parachutes/python:3.12 | |
| run_command: | |
| - pip install --upgrade setuptools wheel | |
| - pip install 'numpy>=1.23' 'onnxruntime-gpu>=1.16' 'nvidia-cudnn-cu12>=9.0' 'nvidia-cublas-cu12>=12.8' 'nvidia-cuda-runtime-cu12>=12.8' 'opencv-python>=4.7' 'pillow>=9.5' 'huggingface_hub>=0.19.4' 'pydantic>=2.0' 'pyyaml>=6.0' 'aiohttp>=3.9' 'torch>=2.8' | |
| NodeSelector: | |
| gpu_count: 1 | |
| min_vram_gb_per_gpu: 16 | |
| # max_hourly_price_per_gpu belongs in NodeSelector, not Chute. | |
| # pro_6000 lists at $1.80/hr on api.chutes.ai/pricing; 2.00 leaves headroom. | |
| max_hourly_price_per_gpu: 2.00 | |
| # TEE mandate for new SN44 chutes: include=['pro_6000'] + Chute.tee=true. | |
| include: | |
| - "pro_6000" | |
| # Chutes caps exclude at 5. Pattern from production person/vehicle chute. | |
| exclude: | |
| - "5090" | |
| - "b200" | |
| - "h200" | |
| - "mi300x" | |
| Chute: | |
| # Chute.__init__ only accepts these params; anything else is silently | |
| # dropped by safe_instantiate() in the template. Previously-used keys | |
| # `shutdown_after` (was dropped because the real name is | |
| # `shutdown_after_seconds`) and `timeout_seconds` (not a Chute param) | |
| # caused the chute to auto-deprovision after 5 min of idle. | |
| concurrency: 4 | |
| max_instances: 5 | |
| scaling_threshold: 0.5 | |
| shutdown_after_seconds: 288000 # 80h idle tolerance | |
| # SDK source has Chute(tee: bool = False) — required for new SN44 chutes. | |
| tee: true | |