def hyperparameters(self):
"""Return hyperparameters used by your custom TensorFlow code during model training."""
hyperparameters = super(TensorFlow, self).hyperparameters()
self.checkpoint_path = self.checkpoint_path or self._default_s3_path(
'checkpoints')
if self._script_mode_enabled():
self.model_dir = self.model_dir or self._default_s3_path('model')
additional_hyperparameters = {'model_dir': self.model_dir}
if 'parameter_server' in self.distributions:
enabled = self.distributions['parameter_server'].get(
'enabled', False)
additional_hyperparameters[self.LAUNCH_PS_ENV_NAME] = enabled
else:
additional_hyperparameters = {
'checkpoint_path': self.checkpoint_path,
'training_steps': self.training_steps,
'evaluation_steps': self.evaluation_steps,
'sagemaker_requirements': self.requirements_file
}
hyperparameters.update(
Framework._json_encode_hyperparameters(additional_hyperparameters))
return hyperparameters
def hyperparameters(self):
"""Return hyperparameters used by your custom TensorFlow code during model training."""
hyperparameters = super(TensorFlow, self).hyperparameters()
self.checkpoint_path = self.checkpoint_path or self._default_s3_path('checkpoints')
mpi_enabled = False
if self._script_mode_enabled():
additional_hyperparameters = {}
if 'parameter_server' in self.distributions:
ps_enabled = self.distributions['parameter_server'].get('enabled', False)
additional_hyperparameters[self.LAUNCH_PS_ENV_NAME] = ps_enabled
if 'mpi' in self.distributions:
mpi_dict = self.distributions['mpi']
mpi_enabled = mpi_dict.get('enabled', False)
additional_hyperparameters[self.LAUNCH_MPI_ENV_NAME] = mpi_enabled
additional_hyperparameters[self.MPI_NUM_PROCESSES_PER_HOST] = mpi_dict.get('processes_per_host', 1)
additional_hyperparameters[self.MPI_CUSTOM_MPI_OPTIONS] = mpi_dict.get('custom_mpi_options', '')
self.model_dir = self.model_dir or self._default_s3_path('model', mpi=mpi_enabled)
additional_hyperparameters['model_dir'] = self.model_dir
else:
additional_hyperparameters = {'checkpoint_path': self.checkpoint_path,
'training_steps': self.training_steps,
'evaluation_steps': self.evaluation_steps,
'sagemaker_requirements': self.requirements_file}
hyperparameters.update(Framework._json_encode_hyperparameters(additional_hyperparameters))
return hyperparameters
def hyperparameters(self):
"""Return hyperparameters used by your custom TensorFlow code during model training."""
hyperparameters = super(TensorFlow, self).hyperparameters()
additional_hyperparameters = {}
if "parameter_server" in self.distribution:
ps_enabled = self.distribution["parameter_server"].get(
"enabled", False)
additional_hyperparameters[self.LAUNCH_PS_ENV_NAME] = ps_enabled
mpi_enabled = False
if "mpi" in self.distribution:
mpi_dict = self.distribution["mpi"]
mpi_enabled = mpi_dict.get("enabled", False)
additional_hyperparameters[self.LAUNCH_MPI_ENV_NAME] = mpi_enabled
if mpi_dict.get("processes_per_host"):
additional_hyperparameters[
self.MPI_NUM_PROCESSES_PER_HOST] = mpi_dict.get(
"processes_per_host")
additional_hyperparameters[
self.MPI_CUSTOM_MPI_OPTIONS] = mpi_dict.get(
"custom_mpi_options", "")
if self.model_dir is not False:
self.model_dir = self.model_dir or self._default_s3_path(
"model", mpi=mpi_enabled)
additional_hyperparameters["model_dir"] = self.model_dir
hyperparameters.update(
Framework._json_encode_hyperparameters(additional_hyperparameters))
return hyperparameters
def hyperparameters(self):
"""Return hyperparameters used by your custom PyTorch code during model training."""
hyperparameters = super(HuggingFace, self).hyperparameters()
additional_hyperparameters = self._distribution_configuration(
distribution=self.distribution
)
hyperparameters.update(Framework._json_encode_hyperparameters(additional_hyperparameters))
return hyperparameters
def hyperparameters(self):
"""Return hyperparameters used by your custom PyTorch code during model training."""
hyperparameters = super(HuggingFace, self).hyperparameters()
distributed_training_hyperparameters = self._distribution_configuration(
distribution=self.distribution)
hyperparameters.update(
Framework._json_encode_hyperparameters(
distributed_training_hyperparameters))
if self.compiler_config:
training_compiler_hyperparameters = self.compiler_config._to_hyperparameter_dict(
)
hyperparameters.update(
Framework._json_encode_hyperparameters(
training_compiler_hyperparameters))
return hyperparameters
def hyperparameters(self):
"""Return hyperparameters used by your custom TensorFlow code during model training."""
hyperparameters = super(RLEstimator, self).hyperparameters()
additional_hyperparameters = {
SAGEMAKER_OUTPUT_LOCATION: self.output_path,
# TODO: can be applied to all other estimators
SAGEMAKER_ESTIMATOR: SAGEMAKER_ESTIMATOR_VALUE,
}
hyperparameters.update(Framework._json_encode_hyperparameters(additional_hyperparameters))
return hyperparameters
def hyperparameters(self):
"""Return hyperparameters used by your custom Chainer code during training."""
hyperparameters = super(Chainer, self).hyperparameters()
additional_hyperparameters = {Chainer._use_mpi: self.use_mpi,
Chainer._num_processes: self.num_processes,
Chainer._process_slots_per_host: self.process_slots_per_host,
Chainer._additional_mpi_options: self.additional_mpi_options}
# remove unset keys.
additional_hyperparameters = {k: v for k, v in additional_hyperparameters.items() if v}
hyperparameters.update(Framework._json_encode_hyperparameters(additional_hyperparameters))
return hyperparameters
def hyperparameters(self):
"""Return hyperparameters used by your custom Chainer code during training."""
hyperparameters = super(Chainer, self).hyperparameters()
additional_hyperparameters = {Chainer._use_mpi: self.use_mpi,
Chainer._num_processes: self.num_processes,
Chainer._process_slots_per_host: self.process_slots_per_host,
Chainer._additional_mpi_options: self.additional_mpi_options}
# remove unset keys.
additional_hyperparameters = {k: v for k, v in additional_hyperparameters.items() if v}
hyperparameters.update(Framework._json_encode_hyperparameters(additional_hyperparameters))
return hyperparameters
def hyperparameters(self):
"""Return hyperparameters used by your custom TensorFlow code during model training."""
hyperparameters = super(TensorFlow, self).hyperparameters()
additional_hyperparameters = {}
if "parameter_server" in self.distribution:
ps_enabled = self.distribution["parameter_server"].get(
"enabled", False)
additional_hyperparameters[self.LAUNCH_PS_ENV_NAME] = ps_enabled
mpi_enabled = False
if "mpi" in self.distribution:
mpi_dict = self.distribution["mpi"]
mpi_enabled = mpi_dict.get("enabled", False)
additional_hyperparameters[self.LAUNCH_MPI_ENV_NAME] = mpi_enabled
if mpi_dict.get("processes_per_host"):
additional_hyperparameters[
self.MPI_NUM_PROCESSES_PER_HOST] = mpi_dict.get(
"processes_per_host")
additional_hyperparameters[
self.MPI_CUSTOM_MPI_OPTIONS] = mpi_dict.get(
"custom_mpi_options", "")
if fw.get_mp_parameters(self.distribution):
additional_hyperparameters[
"mp_parameters"] = fw.get_mp_parameters(self.distribution)
elif "modelparallel" in self.distribution.get("smdistributed", {}):
raise ValueError(
"Cannot use Model Parallelism without MPI enabled!")
if "smdistributed" in self.distribution:
# smdistributed strategy selected
smdistributed = self.distribution["smdistributed"]
smdataparallel_enabled = smdistributed.get("dataparallel", {}).get(
"enabled", False)
additional_hyperparameters[
self.LAUNCH_SM_DDP_ENV_NAME] = smdataparallel_enabled
additional_hyperparameters[self.INSTANCE_TYPE] = self.instance_type
if self.model_dir is not False:
self.model_dir = self.model_dir or self._default_s3_path(
"model", mpi=mpi_enabled)
additional_hyperparameters["model_dir"] = self.model_dir
hyperparameters.update(
Framework._json_encode_hyperparameters(additional_hyperparameters))
return hyperparameters
def hyperparameters(self):
"""Return hyperparameters used by your custom TensorFlow code during model training."""
hyperparameters = super(TensorFlow, self).hyperparameters()
additional_hyperparameters = self._distribution_configuration(
self.distribution)
if self.model_dir is not False:
self.model_dir = self.model_dir or self._default_s3_path(
"model",
mpi=additional_hyperparameters.get(self.LAUNCH_MPI_ENV_NAME,
False))
additional_hyperparameters["model_dir"] = self.model_dir
hyperparameters.update(
Framework._json_encode_hyperparameters(additional_hyperparameters))
return hyperparameters
def validate_estimator_customparams(estimator: Framework,
mock_session: Session, descriptor,
sagemaker_config: SageMakerExecutorConfig):
assert estimator.metric_definitions == get_metric_definitions(descriptor)
assert estimator.hyperparameters()
assert estimator.source_dir == SOURCE_DIR
assert estimator.entry_point == ScriptSourceDirectory.PYTHON_ENTRY_POINT
assert estimator.sagemaker_session == mock_session
assert estimator.image_name == descriptor.env.docker_image
assert estimator.framework_version == descriptor.ml.framework_version
assert estimator.train_instance_type == descriptor.hardware.instance_type
assert estimator.train_instance_count == descriptor.hardware.distributed.num_instances
assert estimator.role == sagemaker_config.sm_role
assert estimator.output_path == f"s3://{sagemaker_config.s3_output_bucket}"
assert estimator.security_group_ids == sagemaker_config.security_group_ids
assert estimator.subnets == sagemaker_config.subnets
def hyperparameters(self):
"""Return hyperparameters used by your custom TensorFlow code during model training."""
hyperparameters = super(TensorFlow, self).hyperparameters()
self.checkpoint_path = self.checkpoint_path or self._default_s3_path(
"checkpoints")
mpi_enabled = False
if self._script_mode_enabled():
additional_hyperparameters = {}
if "parameter_server" in self.distributions:
ps_enabled = self.distributions["parameter_server"].get(
"enabled", False)
additional_hyperparameters[
self.LAUNCH_PS_ENV_NAME] = ps_enabled
if "mpi" in self.distributions:
mpi_dict = self.distributions["mpi"]
mpi_enabled = mpi_dict.get("enabled", False)
additional_hyperparameters[
self.LAUNCH_MPI_ENV_NAME] = mpi_enabled
if mpi_dict.get("processes_per_host"):
additional_hyperparameters[
self.MPI_NUM_PROCESSES_PER_HOST] = mpi_dict.get(
"processes_per_host")
additional_hyperparameters[
self.MPI_CUSTOM_MPI_OPTIONS] = mpi_dict.get(
"custom_mpi_options", "")
self.model_dir = self.model_dir or self._default_s3_path(
"model", mpi=mpi_enabled)
additional_hyperparameters["model_dir"] = self.model_dir
else:
additional_hyperparameters = {
"checkpoint_path": self.checkpoint_path,
"training_steps": self.training_steps,
"evaluation_steps": self.evaluation_steps,
"sagemaker_requirements": self.requirements_file,
}
hyperparameters.update(
Framework._json_encode_hyperparameters(additional_hyperparameters))
return hyperparameters
def hyperparameters(self):
"""Return hyperparameters used by your custom TensorFlow code during model training."""
hyperparameters = super(MXNet, self).hyperparameters()
if not self.checkpoint_path:
local_code = get_config_value('local.local_code', self.sagemaker_session.config)
if self.sagemaker_session.local_mode and local_code:
self.checkpoint_path = '/opt/ml/shared/checkpoints'
else:
self.checkpoint_path = os.path.join(self. ,
self._current_job_name, 'checkpoints')
additional_hyperparameters = {'checkpoint_path': self.checkpoint_path}
hyperparameters.update(Framework._json_encode_hyperparameters(additional_hyperparameters))
return hyperparameters
def hyperparameters(self):
"""Return hyperparameters used by your custom TensorFlow code during model training."""
hyperparameters = super(TensorFlow, self).hyperparameters()
if not self.checkpoint_path:
self.checkpoint_path = os.path.join(self.output_path,
self._current_job_name,
'checkpoints')
additional_hyperparameters = {
'checkpoint_path': self.checkpoint_path,
'training_steps': self.training_steps,
'evaluation_steps': self.evaluation_steps
}
hyperparameters.update(
Framework._json_encode_hyperparameters(additional_hyperparameters))
return hyperparameters
def hyperparameters(self):
"""Return hyperparameters used by your custom TensorFlow code during model training."""
hyperparameters = super(TensorFlow, self).hyperparameters()
if not self.checkpoint_path:
local_code = get_config_value('local.local_code', self.sagemaker_session.config)
if self.sagemaker_session.local_mode and local_code:
self.checkpoint_path = '/opt/ml/shared/checkpoints'
else:
self.checkpoint_path = os.path.join(self.output_path,
self._current_job_name, 'checkpoints')
additional_hyperparameters = {'checkpoint_path': self.checkpoint_path,
'training_steps': self.training_steps,
'evaluation_steps': self.evaluation_steps,
'sagemaker_requirements': self.requirements_file}
hyperparameters.update(Framework._json_encode_hyperparameters(additional_hyperparameters))
return hyperparameters
def test_attach_no_job_name_framework(sagemaker_session):
with pytest.raises(ValueError) as error:
Framework.attach(training_job_name=None,
sagemaker_session=sagemaker_session)
assert 'must specify training_job name' in str(error)
