Skip to main content

detect_sm_for_device

bge_m3_embedding_server::embedder::sm_detect

Function detect_sm_for_device 

Source 
pub(crate) fn detect_sm_for_device(device_id: u32) -> Option<String>
Expand description

Returns the GPU compute capability as a smXY string for the given CUDA device, or None if detection fails for any reason.

The format is the same one ORT uses in its engine plan basenames (e.g. _sm120.engine), so the return value can be plugged directly into the SM-filtered enumerators in super::trt_cache.

This function is a thin wrapper around nvidia-smi. It is intended to be called once per worker at TRT prewarm time and the result cached; do not call it on every request.

Failure-mode contract:

  • nvidia-smi binary missing or unexecutable → None
  • subprocess exits non-zero → None
  • stdout cannot be parsed as "X.Y" → None

All failure cases are non-panicking. The caller is expected to log a WARN and proceed with None semantics (no SM filter applied).