InferenceError — the typed error surface that flows up to the RequestActor regardless of whether the bottleneck was GPU memory, GIL contention, or remote provider quota (doc §6.2).
InferenceError
RequestActor