def _control_loop(self) -> None:
assert self.workloads is not None
for wkld, response_func in self.workloads:
logging.debug(f"Received wkld {wkld.kind}.")
try:
if wkld.kind == workload.Workload.Kind.RUN_STEP:
# Configure the state for a training step.
self.train_response_func = response_func
self.train_workload_batches = 0
self.train_workload_inputs = 0
self.train_workload_metrics = []
self.train_workload_len = wkld.num_batches
self.multiplexer.set_batches_requested(wkld.num_batches)
return
elif wkld.kind == workload.Workload.Kind.COMPUTE_VALIDATION_METRICS:
action = "validation"
response = {
"metrics": self._compute_validation_metrics(),
"stop_requested": self.context.get_stop_requested(),
} # type: workload.Response
elif wkld.kind == workload.Workload.Kind.CHECKPOINT_MODEL:
action = "checkpointing"
if self.is_chief:
metadata = {
"determined_version": det.__version__,
"steps_completed": self.steps_completed,
"framework": f"tensorflow-{tf.__version__}",
"format": "saved_weights",
}
with self.context._core.checkpoint.store_path(metadata) as (
path,
storage_id,
):
self._save_checkpoint(path)
response = {"uuid": storage_id}
else:
response = {}
else:
raise AssertionError(f"Unknown workload kind {wkld.kind}.")
except det.InvalidHP as e:
logging.info(f"Invalid hyperparameter exception during {action}: {e}")
response = workload.InvalidHP()
response_func(response)
self.upload_tb_files()
# End-of-training.
self.multiplexer._corrected_train_end()
raise det.errors.WorkerFinishedGracefully()
def _run(self) -> None:
assert self.workloads is not None
for w, response_func in self.workloads:
try:
if w.kind == workload.Workload.Kind.RUN_STEP:
action = "training"
response = {
"metrics":
self._train_for_step(
w.step_id,
w.num_batches,
w.total_batches_processed,
),
"stop_requested":
self.context.get_stop_requested(),
} # type: workload.Response
elif w.kind == workload.Workload.Kind.COMPUTE_VALIDATION_METRICS:
action = "validation"
response = {
"metrics": self._compute_validation_metrics(),
"stop_requested": self.context.get_stop_requested(),
}
elif w.kind == workload.Workload.Kind.CHECKPOINT_MODEL:
action = "checkpointing"
if self.is_chief:
metadata = {
"steps_completed": self.steps_completed,
"framework": f"torch-{torch.__version__}",
"format": "pickle",
}
with self.context._core.checkpoint.store_path(
metadata) as (
path,
storage_id,
):
self._save(path)
response = {"uuid": storage_id}
else:
response = {}
else:
raise AssertionError("Unexpected workload: {}".format(
w.kind))
except det.InvalidHP as e:
logging.info(
f"Invalid hyperparameter exception during {action}: {e}")
response = workload.InvalidHP()
response_func(response)
def control_loop(self) -> None:
core = self.estimator_trial_controller.context._core
assert self.estimator_trial_controller.workloads is not None
for wkld, response_func in self.estimator_trial_controller.workloads:
logging.debug(f"Received wkld {wkld.kind}.")
try:
if wkld.kind == workload.Workload.Kind.RUN_STEP:
# Store values for the training loop.
self.num_batches = wkld.num_batches
self.train_response_func = response_func
# Break out of the control loop so that the train process
# re-enters the train_and_evaluate() loop.
return
elif wkld.kind == workload.Workload.Kind.COMPUTE_VALIDATION_METRICS:
action = "validation"
response = {
"metrics": self._compute_validation_metrics(),
"stop_requested": (
self.estimator_trial_controller.context.get_stop_requested()
),
} # type: workload.Response
elif wkld.kind == workload.Workload.Kind.CHECKPOINT_MODEL:
action = "checkpointing"
self._save_model()
if self.estimator_trial_controller.is_chief:
metadata = {
"steps_completed": self.steps_completed,
"framework": f"tensorflow-{tf.__version__}",
"format": "saved_model",
}
with core.checkpoint.store_path(metadata) as (path, storage_id):
self._checkpoint_model(path)
response = {"uuid": storage_id}
else:
response = {}
else:
raise AssertionError(f"Unknown wkld kind {wkld.kind}.")
except det.InvalidHP as e:
logging.info(f"Invalid hyperparameter exception during {action}: {e}")
response = workload.InvalidHP()
response_func(response)
# End-of-training.
raise det.errors.WorkerFinishedGracefully("Exiting normally.")
def _run(self) -> None:
assert self.workloads is not None
for w, response_func in self.workloads:
try:
if w.kind == workload.Workload.Kind.RUN_STEP:
action = "training"
response = {
"metrics":
self._train_for_step(
w.step_id,
w.num_batches,
w.total_batches_processed,
),
"stop_requested":
self.context.get_stop_requested(),
} # type: workload.Response
elif w.kind == workload.Workload.Kind.COMPUTE_VALIDATION_METRICS:
action = "validation"
response = {
"metrics": self._compute_validation_metrics(),
"stop_requested": self.context.get_stop_requested(),
}
elif w.kind == workload.Workload.Kind.CHECKPOINT_MODEL:
action = "checkpointing"
# The checkpointing api would have been sufficient if the base_path for the
# storage manager is guaranteed to be a shared file system.
#
# Since we can't guarantee that, we use the base storage_manager instead for
# more flexibility. Since checkpoints can be distributed across multiple
# nodes, we will use the same uuid and separate path but each node
# will upload its checkpoints to the storage manager individually.
storage_manager = self.context._core.checkpoint._storage_manager
if self.is_chief:
metadata = {
"steps_completed": self.steps_completed,
"framework": f"torch-{torch.__version__}",
"format": "pickle",
}
storage_id = str(uuid.uuid4())
with storage_manager.store_path(storage_id) as path:
# Broadcast checkpoint path to all ranks.
self.context.distributed.broadcast(
(storage_id, path))
self._save(path)
# Gather resources across nodes.
all_resources = self.context.distributed.gather(
storage.StorageManager._list_directory(path))
resources = {
k: v
for d in all_resources for k, v in d.items()
}
self.context._core.checkpoint._report_checkpoint(
storage_id, resources, metadata)
response = {"uuid": storage_id}
else:
storage_id, path = self.context.distributed.broadcast(
None)
self._save(path)
# Gather resources across nodes.
_ = self.context.distributed.gather(
storage.StorageManager._list_directory(path))
if self.context.distributed.local_rank == 0:
storage_manager.post_store_path(
str(path), storage_id)
response = {}
else:
raise AssertionError("Unexpected workload: {}".format(
w.kind))
except det.InvalidHP as e:
logging.info(
f"Invalid hyperparameter exception during {action}: {e}")
response = workload.InvalidHP()
response_func(response)
self.upload_tb_files()
