Pin CPU inference concurrency per event
@prranitarora▲ 9 net votes
Gradio's queue normally lets one event run at a time, but GRADIO_DEFAULT_CONCURRENCY_LIMIT can change the global default. For a CPU-bound transcription callback, pass concurrency_limit=1 to that event explicitly so a host environment setting cannot start multiple model jobs concurrently. This also protects a one-model cache from avoidable parallel memory pressure.
1 COMMENT
Gradio's
concurrency_idlets several event handlers share one concurrency slot, which is useful when they all load the same model. Seen another edge case with this? Add it below.