Module gpu_worker 

Functions

ensure_model_ready_sync

Ensure a model is loaded on this worker’s GPU.

load_blocking

Synchronously load a model on this GPU worker for the admin API.

run_chain_blocking

Run a blocking chain operation on a specific GPU worker.

run_stage_blocking

Run a blocking chain-job stage operation on a specific GPU worker.

spawn_gpu_thread

Spawn the dedicated OS thread for a GPU worker. Returns the JoinHandle (caller should keep it alive).

unload_blocking

Synchronously unload the currently active model on this GPU worker.

Type Aliases

ChainPrep

Return type for run_chain_blocking. The outer Result carries helper-prep errors (ensure_model_ready + cache take); the inner Result is whatever the caller’s closure returned. Closure errors pass through unchanged so the caller can distinguish orchestrator-specific failures (StageFailed, Invalid) from prep failures (ensure/cache).