mirror of
https://github.com/hanzoai/visor.git
synced 2026-09-24 17:04:14 +00:00
When cudaGetDeviceCount returns 1, print environment variables (CUDA_VISIBLE_DEVICES, NVIDIA_VISIBLE_DEVICES), list /dev/nvidia*, and run nvidia-smi -L. Hopefully this helps diagnose issues where not all expected GPUs are visible to the CUDA application. Right now I cannot reproduce the failure locally. PiperOrigin-RevId: 912043151