MCPcopy Create free account
hub / github.com/AI-Hypercomputer/maxtext / wait_for_all_slices

Function wait_for_all_slices

src/MaxText/elastic_train.py:321–330  ·  view source on GitHub ↗
(elastic_manager: manager.Manager)

Source from the content-addressed store, hash-verified

319
320
321def wait_for_all_slices(elastic_manager: manager.Manager) -> None:
322 elastic_manager.good_slice_indices = elastic_manager.get_slice_availability()
323 while len(elastic_manager.good_slice_indices) < elastic_manager.total_slice_count:
324 max_logging.log(
325 f"Only {elastic_manager.good_slice_count} slices out of {elastic_manager.total_slice_count} available. "
326 "Sleeping for 5 seconds."
327 )
328 time.sleep(5)
329 elastic_manager.good_slice_indices = elastic_manager.get_slice_availability()
330 max_logging.log("All slices are available")
331
332
333def elastic_initialize(devices: Sequence[jax.Device]) -> manager.Manager:

Callers 1

elastic_initializeFunction · 0.85

Calls

no outgoing calls

Tested by

no test coverage detected