Run Experiment

Deploy and run ML experiment: $ARGUMENTS

Workflow

Step 0: Compute Resource Guard (MANDATORY)

Before doing ANYTHING else, run /aris-compute-guard to verify that compute resources are actually available.

If /aris-compute-guard is not available as a sub-skill, perform the check inline:

1. Local GPU (Linux): Run nvidia-smi --query-gpu=index,memory.used,memory.total --format=csv,noheader. A GPU is free if memory.used < 500 MiB.
2. Local GPU (Mac): Run python3 -c "import torch; print(torch.backends.mps.is_available())".
3. Remote server: Run ssh <server> nvidia-smi --query-gpu=index,memory.used,memory.total --format=csv,noheader.
4. Modal: Check modal token verify — Modal is serverless and always available if configured.