Use NVML directly and cache nvidia-smi calls - #114

Open

Qubitium wants to merge 3 commits into

mainfrom

fix/nvml-nvidia-smi-cache

Commits on Aug 13, 2026

feat(nvidia): use NVML when present, cache nvidia-smi with configurable TTL
Qubitium
committed
a6d9903
refactor: rename Device.settings to Device.config, default nvidia-smi cache TTL to 0, ensure GIL=0 thread safety
Qubitium
committed
be11838
chore: bump version to 0.5.7
Qubitium
committed
0febcea