Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add supported inference and incremental training configs #4637

Merged
merged 10 commits into from
May 2, 2024
Merged

Add supported inference and incremental training configs #4637

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
a0fd751
supported inference configs
Captainia May 1, 2024
7f08766
add tests
Captainia May 1, 2024
1a1a6bb
format
Captainia May 1, 2024
edf61af
tests
Captainia May 1, 2024
4ec00fd
tests
Captainia May 1, 2024
6340736
address comments
Captainia May 1, 2024
dc33402
format and address comments
Captainia May 1, 2024
096de7b
updates
Captainia May 1, 2024
fd720d8
formt
Captainia May 1, 2024
cf0e8ab
format
Captainia May 2, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
27 changes: 15 additions & 12 deletions src/sagemaker/jumpstart/estimator.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -112,7 +112,6 @@ def __init__(
disable_output_compression: Optional[bool] = None,
enable_remote_debug: Optional[Union[bool, PipelineVariable]] = None,
config_name: Optional[str] = None,
inference_config_name: Optional[str] = None,
):
"""Initializes a ``JumpStartEstimator``.

Expand Down Expand Up @@ -506,9 +505,6 @@ def __init__(
Specifies whether RemoteDebug is enabled for the training job
config_name (Optional[str]):
Name of the training configuration to apply to the Estimator. (Default: None).
inference_config_name (Optional[str]):
Name of the inference configuraion to apply to the Estimator,
to be used when deploying the fine-tuned mode. (Default: None).

Raises:
ValueError: If the model ID is not recognized by JumpStart.
Expand Down Expand Up @@ -587,8 +583,7 @@ def _validate_model_id_and_get_type_hook():
disable_output_compression=disable_output_compression,
enable_infra_check=enable_infra_check,
enable_remote_debug=enable_remote_debug,
training_config_name=config_name,
inference_config_name=inference_config_name,
config_name=config_name,
)

self.model_id = estimator_init_kwargs.model_id
Expand All @@ -602,8 +597,7 @@ def _validate_model_id_and_get_type_hook():
self.role = estimator_init_kwargs.role
self.sagemaker_session = estimator_init_kwargs.sagemaker_session
self._enable_network_isolation = estimator_init_kwargs.enable_network_isolation
self.training_config_name = estimator_init_kwargs.training_config_name
self.inference_config_name = estimator_init_kwargs.inference_config_name
self.config_name = estimator_init_kwargs.config_name
self.init_kwargs = estimator_init_kwargs.to_kwargs_dict(False)

super(JumpStartEstimator, self).__init__(**estimator_init_kwargs.to_kwargs_dict())
Expand Down Expand Up @@ -679,7 +673,7 @@ def fit(
tolerate_vulnerable_model=self.tolerate_vulnerable_model,
tolerate_deprecated_model=self.tolerate_deprecated_model,
sagemaker_session=self.sagemaker_session,
config_name=self.training_config_name,
config_name=self.config_name,
)

return super(JumpStartEstimator, self).fit(**estimator_fit_kwargs.to_kwargs_dict())
Expand All @@ -692,6 +686,7 @@ def attach(
model_version: Optional[str] = None,
sagemaker_session: session.Session = DEFAULT_JUMPSTART_SAGEMAKER_SESSION,
model_channel_name: str = "model",
config_name: Optional[str] = None,
) -> "JumpStartEstimator":
"""Attach to an existing training job.

Expand Down Expand Up @@ -727,6 +722,8 @@ def attach(
model data will be downloaded (default: 'model'). If no channel
with the same name exists in the training job, this option will
be ignored.
config_name (str): Optional. Name of the training configuration to use
when attaching to the training job. (Default: None).

Returns:
Instance of the calling ``JumpStartEstimator`` Class with the attached
Expand All @@ -738,7 +735,6 @@ def attach(
"""
config_name = None
if model_id is None:

model_id, model_version, _, config_name = get_model_info_from_training_job(
training_job_name=training_job_name, sagemaker_session=sagemaker_session
)
Expand All @@ -752,6 +748,9 @@ def attach(
"tolerate_deprecated_model": True, # model is already trained
}

if config_name:
additional_kwargs.update({"config_name": config_name})

model_specs = verify_model_region_and_return_specs(
model_id=model_id,
version=model_version,
Expand Down Expand Up @@ -810,6 +809,7 @@ def deploy(
dependencies: Optional[List[str]] = None,
git_config: Optional[Dict[str, str]] = None,
use_compiled_model: bool = False,
inference_config_name: Optional[str] = None,
) -> PredictorBase:
"""Creates endpoint from training job.

Expand Down Expand Up @@ -1045,6 +1045,8 @@ def deploy(
(Default: None).
use_compiled_model (bool): Flag to select whether to use compiled
(optimized) model. (Default: False).
inference_config_name (Optional[str]): Name of the inference configuration to
be used in the model. (Default: None).
"""
self.orig_predictor_cls = predictor_cls

Expand Down Expand Up @@ -1097,7 +1099,8 @@ def deploy(
git_config=git_config,
use_compiled_model=use_compiled_model,
training_instance_type=self.instance_type,
config_name=self.inference_config_name,
training_config_name=self.config_name,
inference_config_name=inference_config_name,
)

predictor = super(JumpStartEstimator, self).deploy(
Expand All @@ -1114,7 +1117,7 @@ def deploy(
tolerate_deprecated_model=self.tolerate_deprecated_model,
tolerate_vulnerable_model=self.tolerate_vulnerable_model,
sagemaker_session=self.sagemaker_session,
config_name=self.inference_config_name,
config_name=estimator_deploy_kwargs.config_name,
)

# If a predictor class was passed, do not mutate predictor
Expand Down
94 changes: 24 additions & 70 deletions src/sagemaker/jumpstart/factory/estimator.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -131,8 +131,7 @@ def get_init_kwargs(
disable_output_compression: Optional[bool] = None,
enable_infra_check: Optional[Union[bool, PipelineVariable]] = None,
enable_remote_debug: Optional[Union[bool, PipelineVariable]] = None,
training_config_name: Optional[str] = None,
inference_config_name: Optional[str] = None,
config_name: Optional[str] = None,
) -> JumpStartEstimatorInitKwargs:
"""Returns kwargs required to instantiate `sagemaker.estimator.Estimator` object."""

Expand Down Expand Up @@ -191,8 +190,7 @@ def get_init_kwargs(
disable_output_compression=disable_output_compression,
enable_infra_check=enable_infra_check,
enable_remote_debug=enable_remote_debug,
training_config_name=training_config_name,
inference_config_name=inference_config_name,
config_name=config_name,
)

estimator_init_kwargs = _add_model_version_to_kwargs(estimator_init_kwargs)
Expand Down Expand Up @@ -295,7 +293,8 @@ def get_deploy_kwargs(
use_compiled_model: Optional[bool] = None,
model_name: Optional[str] = None,
training_instance_type: Optional[str] = None,
config_name: Optional[str] = None,
training_config_name: Optional[str] = None,
inference_config_name: Optional[str] = None,
) -> JumpStartEstimatorDeployKwargs:
"""Returns kwargs required to call `deploy` on `sagemaker.estimator.Estimator` object."""

Expand Down Expand Up @@ -323,7 +322,8 @@ def get_deploy_kwargs(
tolerate_vulnerable_model=tolerate_vulnerable_model,
tolerate_deprecated_model=tolerate_deprecated_model,
sagemaker_session=sagemaker_session,
config_name=config_name,
training_config_name=training_config_name,
config_name=inference_config_name,
)

model_init_kwargs: JumpStartModelInitKwargs = model.get_init_kwargs(
Expand Down Expand Up @@ -352,7 +352,7 @@ def get_deploy_kwargs(
tolerate_deprecated_model=tolerate_deprecated_model,
training_instance_type=training_instance_type,
disable_instance_type_logging=True,
config_name=config_name,
config_name=model_deploy_kwargs.config_name,
)

estimator_deploy_kwargs: JumpStartEstimatorDeployKwargs = JumpStartEstimatorDeployKwargs(
Expand Down Expand Up @@ -397,7 +397,7 @@ def get_deploy_kwargs(
tolerate_vulnerable_model=model_init_kwargs.tolerate_vulnerable_model,
tolerate_deprecated_model=model_init_kwargs.tolerate_deprecated_model,
use_compiled_model=use_compiled_model,
config_name=config_name,
config_name=model_deploy_kwargs.config_name,
)

return estimator_deploy_kwargs
Expand Down Expand Up @@ -453,7 +453,7 @@ def _add_instance_type_and_count_to_kwargs(
tolerate_deprecated_model=kwargs.tolerate_deprecated_model,
tolerate_vulnerable_model=kwargs.tolerate_vulnerable_model,
sagemaker_session=kwargs.sagemaker_session,
config_name=kwargs.training_config_name,
config_name=kwargs.config_name,
)

kwargs.instance_count = kwargs.instance_count or 1
Expand All @@ -477,15 +477,15 @@ def _add_tags_to_kwargs(kwargs: JumpStartEstimatorInitKwargs) -> JumpStartEstima
tolerate_vulnerable_model=kwargs.tolerate_vulnerable_model,
tolerate_deprecated_model=kwargs.tolerate_deprecated_model,
sagemaker_session=kwargs.sagemaker_session,
config_name=kwargs.training_config_name,
config_name=kwargs.config_name,
).version

if kwargs.sagemaker_session.settings.include_jumpstart_tags:
kwargs.tags = add_jumpstart_model_info_tags(
kwargs.tags,
kwargs.model_id,
full_model_version,
config_name=kwargs.training_config_name,
config_name=kwargs.config_name,
scope=JumpStartScriptScope.TRAINING,
)
return kwargs
Expand All @@ -504,7 +504,7 @@ def _add_image_uri_to_kwargs(kwargs: JumpStartEstimatorInitKwargs) -> JumpStartE
tolerate_deprecated_model=kwargs.tolerate_deprecated_model,
tolerate_vulnerable_model=kwargs.tolerate_vulnerable_model,
sagemaker_session=kwargs.sagemaker_session,
config_name=kwargs.training_config_name,
config_name=kwargs.config_name,
)

return kwargs
Expand All @@ -530,7 +530,7 @@ def _add_model_uri_to_kwargs(kwargs: JumpStartEstimatorInitKwargs) -> JumpStartE
sagemaker_session=kwargs.sagemaker_session,
region=kwargs.region,
instance_type=kwargs.instance_type,
config_name=kwargs.training_config_name,
config_name=kwargs.config_name,
)

if (
Expand All @@ -543,7 +543,7 @@ def _add_model_uri_to_kwargs(kwargs: JumpStartEstimatorInitKwargs) -> JumpStartE
tolerate_deprecated_model=kwargs.tolerate_deprecated_model,
tolerate_vulnerable_model=kwargs.tolerate_vulnerable_model,
sagemaker_session=kwargs.sagemaker_session,
config_name=kwargs.training_config_name,
config_name=kwargs.config_name,
)
):
JUMPSTART_LOGGER.warning(
Expand Down Expand Up @@ -579,7 +579,7 @@ def _add_source_dir_to_kwargs(kwargs: JumpStartEstimatorInitKwargs) -> JumpStart
tolerate_vulnerable_model=kwargs.tolerate_vulnerable_model,
region=kwargs.region,
sagemaker_session=kwargs.sagemaker_session,
config_name=kwargs.training_config_name,
config_name=kwargs.config_name,
)

return kwargs
Expand All @@ -600,7 +600,7 @@ def _add_env_to_kwargs(
sagemaker_session=kwargs.sagemaker_session,
script=JumpStartScriptScope.TRAINING,
instance_type=kwargs.instance_type,
config_name=kwargs.training_config_name,
config_name=kwargs.config_name,
)

model_package_artifact_uri = _retrieve_model_package_model_artifact_s3_uri(
Expand All @@ -611,7 +611,7 @@ def _add_env_to_kwargs(
tolerate_deprecated_model=kwargs.tolerate_deprecated_model,
tolerate_vulnerable_model=kwargs.tolerate_vulnerable_model,
sagemaker_session=kwargs.sagemaker_session,
config_name=kwargs.training_config_name,
config_name=kwargs.config_name,
)

if model_package_artifact_uri:
Expand Down Expand Up @@ -639,7 +639,7 @@ def _add_env_to_kwargs(
tolerate_deprecated_model=kwargs.tolerate_deprecated_model,
tolerate_vulnerable_model=kwargs.tolerate_vulnerable_model,
sagemaker_session=kwargs.sagemaker_session,
config_name=kwargs.training_config_name,
config_name=kwargs.config_name,
)
if model_specs.is_gated_model():
raise ValueError(
Expand Down Expand Up @@ -700,7 +700,7 @@ def _add_hyperparameters_to_kwargs(
tolerate_vulnerable_model=kwargs.tolerate_vulnerable_model,
sagemaker_session=kwargs.sagemaker_session,
instance_type=kwargs.instance_type,
config_name=kwargs.training_config_name,
config_name=kwargs.config_name,
)

for key, value in default_hyperparameters.items():
Expand Down Expand Up @@ -734,7 +734,7 @@ def _add_metric_definitions_to_kwargs(
tolerate_vulnerable_model=kwargs.tolerate_vulnerable_model,
sagemaker_session=kwargs.sagemaker_session,
instance_type=kwargs.instance_type,
config_name=kwargs.training_config_name,
config_name=kwargs.config_name,
)
or []
)
Expand Down Expand Up @@ -764,7 +764,7 @@ def _add_estimator_extra_kwargs(
tolerate_deprecated_model=kwargs.tolerate_deprecated_model,
tolerate_vulnerable_model=kwargs.tolerate_vulnerable_model,
sagemaker_session=kwargs.sagemaker_session,
config_name=kwargs.training_config_name,
config_name=kwargs.config_name,
)

for key, value in estimator_kwargs_to_add.items():
Expand Down Expand Up @@ -812,58 +812,12 @@ def _add_config_name_to_kwargs(
tolerate_vulnerable_model=kwargs.tolerate_vulnerable_model,
tolerate_deprecated_model=kwargs.tolerate_deprecated_model,
sagemaker_session=kwargs.sagemaker_session,
config_name=kwargs.training_config_name,
config_name=kwargs.config_name,
)

if kwargs.base_job_name:
_, _, _, base_training_config_name = get_model_info_from_training_job(
training_job_name=kwargs.base_job_name, sagemaker_session=kwargs.sagemaker_session
)

kwargs.training_config_name = (
kwargs.training_config_name
or specs.training_configs.configs.get(
base_training_config_name
).default_incremental_trainig_config
or specs.training_configs.get_top_config_from_ranking().default_incremental_trainig_config # noqa E501 # pylint: disable=c0301
)

if specs.training_configs and specs.training_configs.get_top_config_from_ranking().config_name:
kwargs.training_config_name = (
kwargs.training_config_name
or specs.training_configs.get_top_config_from_ranking().config_name
)

kwargs.inference_config_name = (
kwargs.inference_config_name
or specs.training_configs.configs.get(
kwargs.training_config_name
).default_inference_config
kwargs.config_name = (
kwargs.config_name or specs.training_configs.get_top_config_from_ranking().config_name
)

if (
kwargs.inference_config_name
and kwargs.inference_config_name
not in specs.training_configs.configs.get(
kwargs.training_config_name
).supported_inference_configs
):
raise ValueError(
f"Inference config {kwargs.inference_config_name}"
f"is not supported for model {kwargs.model_id}."
)

if not kwargs.training_config_name:
return kwargs

resolved_config = specs.training_configs.configs[
kwargs.training_config_name
].resolved_config
supported_instance_types = resolved_config.get("supported_training_instance_types", [])
if kwargs.instance_type not in supported_instance_types:
raise ValueError(
f"Instance type {kwargs.instance_type} "
f"is not supported for config {kwargs.training_config_name}."
)

return kwargs
Loading
Loading