Enable async checkpoint by default in Tensorflow model garden.

PiperOrigin-RevId: 532860554

Author: tensorflower-gardener

official/core/train_lib.py

@@ -71,6 +71,7 @@ def __init__(
       controller_cls=orbit.Controller,
       summary_manager: Optional[orbit.utils.SummaryManager] = None,
       eval_summary_manager: Optional[orbit.utils.SummaryManager] = None,
+      enable_async_checkpointing: bool = False,
   ):
     """Constructor.
 
@@ -94,6 +95,8 @@ def __init__(
         summary manager.
       eval_summary_manager: Instance of the eval summary manager to override
         default eval summary manager.
+      enable_async_checkpointing: Optional boolean indicating whether to enable
+        async checkpoint saving.
     """
     self.strategy = distribution_strategy or tf.distribute.get_strategy()
     self._params = params
@@ -115,7 +118,8 @@ def __init__(
         save_summary=save_summary,
         train_actions=train_actions,
         eval_actions=eval_actions,
-        controller_cls=controller_cls)
+        controller_cls=controller_cls,
+        enable_async_checkpointing=enable_async_checkpointing)
 
   @property
   def params(self) -> config_definitions.ExperimentConfig:
@@ -188,13 +192,16 @@ def _maybe_build_checkpoint_manager(
       checkpoint_manager = None
     return checkpoint_manager
 
-  def _build_controller(self,
-                        trainer,
-                        evaluator,
-                        save_summary: bool = True,
-                        train_actions: Optional[List[orbit.Action]] = None,
-                        eval_actions: Optional[List[orbit.Action]] = None,
-                        controller_cls=orbit.Controller) -> orbit.Controller:
+  def _build_controller(
+      self,
+      trainer,
+      evaluator,
+      save_summary: bool = True,
+      train_actions: Optional[List[orbit.Action]] = None,
+      eval_actions: Optional[List[orbit.Action]] = None,
+      controller_cls=orbit.Controller,
+      enable_async_checkpointing: bool = False,
+  ) -> orbit.Controller:
     """Builds a Orbit controler."""
     train_actions = [] if not train_actions else train_actions
     if trainer:
@@ -223,6 +230,7 @@ def _build_controller(self,
         global_step=self.trainer.global_step,
         steps_per_loop=self.params.trainer.steps_per_loop,
         checkpoint_manager=self.checkpoint_manager,
+        enable_async_checkpointing=enable_async_checkpointing,
         summary_dir=os.path.join(self.model_dir, 'train')
         if (save_summary)
         else None,
@@ -309,6 +317,7 @@ def run_experiment(
     controller_cls=orbit.Controller,
     summary_manager: Optional[orbit.utils.SummaryManager] = None,
     eval_summary_manager: Optional[orbit.utils.SummaryManager] = None,
+    enable_async_checkpointing: bool = False,
 ) -> Tuple[tf.keras.Model, Mapping[str, Any]]:
   """Runs train/eval configured by the experiment params.
 
@@ -332,6 +341,8 @@ def run_experiment(
       manager.
     eval_summary_manager: Instance of the eval summary manager to override
       default eval summary manager.
+    enable_async_checkpointing: Optional boolean indicating whether to enable
+        async checkpoint saving.
 
   Returns:
     A 2-tuple of (model, eval_logs).
@@ -353,5 +364,6 @@ def run_experiment(
       controller_cls=controller_cls,
       summary_manager=summary_manager,
       eval_summary_manager=eval_summary_manager,
+      enable_async_checkpointing=enable_async_checkpointing,
   )
   return runner.run()

official/nlp/train.py

@@ -38,6 +38,11 @@
     default=None,
     help='The number of total training steps for the pretraining job.')
 
+flags.DEFINE_bool(
+    'enable_async_checkpointing',
+    default=True,
+    help='A boolean indicating whether to enable async checkpoint saving')
+
 
 def _run_experiment_with_preemption_recovery(params, model_dir):
   """Runs experiment and tries to reconnect when encounting a preemption."""
@@ -62,7 +67,8 @@ def _run_experiment_with_preemption_recovery(params, model_dir):
           task=task,
           mode=FLAGS.mode,
           params=params,
-          model_dir=model_dir)
+          model_dir=model_dir,
+          enable_async_checkpointing=FLAGS.enable_async_checkpointing)
 
       keep_training = False
     except tf.errors.OpError as e:

official/vision/train.py

@@ -32,6 +32,11 @@
 
 FLAGS = flags.FLAGS
 
+flags.DEFINE_bool(
+    'enable_async_checkpointing',
+    default=True,
+    help='A boolean indicating whether to enable async checkpoint saving')
+
 
 def _run_experiment_with_preemption_recovery(params, model_dir):
   """Runs experiment and tries to reconnect when encounting a preemption."""
@@ -60,6 +65,7 @@ def _run_experiment_with_preemption_recovery(params, model_dir):
           eval_summary_manager=summary_manager.maybe_build_eval_summary_manager(
               params=params, model_dir=model_dir
           ),
+          enable_async_checkpointing=FLAGS.enable_async_checkpointing,
       )
 
       keep_training = False