Module startup 

Worker startup: model load, optional cache GC, TRT prewarm, readiness signal.

Structs

StartupOutcome 🔒

Outcome of the blocking startup sequence before the request loop begins.

Functions

detected_sm_for_ep 🔒

Returns the GPU compute capability for TRT EP workers, or None on CPU/CUDA.

startup_worker 🔒

Loads models, runs optional TRT prewarm, and signals readiness.