MCPcopy Index your code
hub / github.com/tensorflow/models / _maybe_save_checkpoint

Method _maybe_save_checkpoint

orbit/controller.py:543–567  ·  view source on GitHub ↗

Conditionally saves a checkpoint. A checkpoint is saved if a `CheckpointManager` is available, and if the required number of steps has elapsed since the last checkpoint was saved (although this condition can be disabled by setting `check_interval=False`). Args: check_interval

(self, check_interval: bool = True)

Source from the content-addressed store, hash-verified

541 self.summary_manager.flush()
542
543 def _maybe_save_checkpoint(self, check_interval: bool = True):
544 """Conditionally saves a checkpoint.
545
546 A checkpoint is saved if a `CheckpointManager` is available, and if the
547 required number of steps has elapsed since the last checkpoint was saved
548 (although this condition can be disabled by setting `check_interval=False`).
549
550 Args:
551 check_interval: Whether to check if the checkpoint interval has fully
552 elapsed. If `False`, a checkpoint is saved regardless of the elapsed
553 steps since the most recent checkpoint, unless no `checkpoint_manager`
554 was provided to `Controller.__init__`.
555
556 Returns:
557 A boolean indicating whether a checkpoint was saved.
558 """
559 if self.checkpoint_manager and self.checkpoint_manager.checkpoint_interval:
560 ckpt_path = self.checkpoint_manager.save(
561 checkpoint_number=self.global_step.numpy(),
562 check_interval=check_interval,
563 options=self._checkpoint_options)
564 if ckpt_path is not None:
565 _log(f"saved checkpoint to {ckpt_path}.")
566 return True
567 return False
568
569 def _require(self, attribute, for_method):
570 """Utility method to raise an error if the given `attribute` is not set."""

Callers 3

trainMethod · 0.95
train_and_evaluateMethod · 0.95
save_checkpointMethod · 0.95

Calls 2

_logFunction · 0.85
saveMethod · 0.80

Tested by

no test coverage detected