File size: 1,178 Bytes
76eef16
 
 
 
 
 
 
 
 
 
 
0bff595
 
76eef16
 
0bff595
76eef16
0bff595
76eef16
 
 
 
 
 
 
 
0bff595
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
Image:
  from_base: parachutes/python:3.12
  run_command:
    - pip install --upgrade setuptools wheel
    # Same runtime as Score's CPU compliance re-run, so the chute and the audit produce the same boxes.
    - pip install 'onnxruntime==1.26.0' 'numpy>=2,<3' 'opencv-python-headless>=4.8' 'pydantic>=2.0' 'pyyaml>=6.0' 'aiohttp>=3.9' 'huggingface_hub>=0.30' 'pillow>=10'
    # The chute template imports torch; the CPU build is enough (the model runs on onnxruntime CPU).
    - pip install torch --index-url https://download.pytorch.org/whl/cpu
  set_workdir: /app

NodeSelector:
  # Chutes runs only TEE (confidential-computing) chutes since 2026-05-12; the cheapest TEE-capable
  # type with capacity is the RTX PRO 6000 ($2.25/h). The price cap keeps us off H200/B200 ($5+/h).
  gpu_count: 1
  min_vram_gb_per_gpu: 16
  max_hourly_price_per_gpu: 2.3
  include:
    - pro_6000

Chute:
  # Tests arrive about once an hour and a cold chute scores 0 (the validator's wake-up call fails),
  # so never shut down for idleness; a keep-warm ping also runs every few minutes.
  shutdown_after_seconds: 86400
  concurrency: 4
  max_instances: 1
  scaling_threshold: 0.75
  tee: true