ob-metaflow-stubs 6.0.3.182rc1__py2.py3-none-any.whl → 6.0.3.183rc0__py2.py3-none-any.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- metaflow-stubs/__init__.pyi +707 -707
- metaflow-stubs/cards.pyi +2 -2
- metaflow-stubs/cli.pyi +2 -2
- metaflow-stubs/cli_components/__init__.pyi +2 -2
- metaflow-stubs/cli_components/utils.pyi +2 -2
- metaflow-stubs/client/__init__.pyi +2 -2
- metaflow-stubs/client/core.pyi +5 -5
- metaflow-stubs/client/filecache.pyi +2 -2
- metaflow-stubs/events.pyi +2 -2
- metaflow-stubs/exception.pyi +2 -2
- metaflow-stubs/flowspec.pyi +5 -5
- metaflow-stubs/generated_for.txt +1 -1
- metaflow-stubs/includefile.pyi +3 -3
- metaflow-stubs/info_file.pyi +2 -2
- metaflow-stubs/metadata_provider/__init__.pyi +2 -2
- metaflow-stubs/metadata_provider/heartbeat.pyi +2 -2
- metaflow-stubs/metadata_provider/metadata.pyi +3 -3
- metaflow-stubs/metadata_provider/util.pyi +2 -2
- metaflow-stubs/metaflow_config.pyi +2 -4
- metaflow-stubs/metaflow_current.pyi +102 -102
- metaflow-stubs/metaflow_git.pyi +2 -2
- metaflow-stubs/mf_extensions/__init__.pyi +2 -2
- metaflow-stubs/mf_extensions/obcheckpoint/__init__.pyi +2 -2
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/__init__.pyi +2 -2
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/__init__.pyi +2 -2
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/card_utils/__init__.pyi +2 -2
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/card_utils/async_cards.pyi +3 -3
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/card_utils/deco_injection_mixin.pyi +2 -2
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/card_utils/extra_components.pyi +3 -3
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/checkpoints/__init__.pyi +2 -2
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/checkpoints/cards/__init__.pyi +2 -2
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/checkpoints/cards/checkpoint_lister.pyi +4 -4
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/checkpoints/cards/lineage_card.pyi +2 -2
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/checkpoints/checkpoint_storage.pyi +4 -4
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/checkpoints/constructors.pyi +2 -2
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/checkpoints/core.pyi +4 -4
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/checkpoints/decorator.pyi +5 -5
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/checkpoints/exceptions.pyi +2 -2
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/checkpoints/final_api.pyi +3 -3
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/checkpoints/lineage.pyi +2 -2
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/datastore/__init__.pyi +2 -2
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/datastore/context.pyi +3 -3
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/datastore/core.pyi +3 -3
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/datastore/decorator.pyi +2 -2
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/datastore/exceptions.pyi +2 -2
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/datastore/task_utils.pyi +3 -3
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/datastore/utils.pyi +2 -2
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/datastructures.pyi +3 -3
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/exceptions.pyi +2 -2
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/hf_hub/__init__.pyi +2 -2
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/hf_hub/decorator.pyi +3 -3
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/modeling_utils/__init__.pyi +2 -2
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/modeling_utils/core.pyi +3 -3
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/modeling_utils/exceptions.pyi +2 -2
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/modeling_utils/model_storage.pyi +4 -4
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/utils/__init__.pyi +2 -2
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/utils/flowspec_utils.pyi +2 -2
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/utils/general.pyi +2 -2
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/utils/identity_utils.pyi +2 -2
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/utils/serialization_handler/__init__.pyi +2 -2
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/utils/serialization_handler/base.pyi +2 -2
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/utils/serialization_handler/tar.pyi +3 -3
- metaflow-stubs/mf_extensions/obcheckpoint/plugins/machine_learning_utilities/utils/tar_utils.pyi +3 -3
- metaflow-stubs/mf_extensions/outerbounds/__init__.pyi +2 -2
- metaflow-stubs/mf_extensions/outerbounds/plugins/__init__.pyi +2 -2
- metaflow-stubs/mf_extensions/outerbounds/plugins/aws/__init__.pyi +2 -2
- metaflow-stubs/mf_extensions/outerbounds/plugins/aws/assume_role_decorator.pyi +2 -2
- metaflow-stubs/mf_extensions/outerbounds/plugins/card_utilities/__init__.pyi +2 -2
- metaflow-stubs/mf_extensions/outerbounds/plugins/card_utilities/async_cards.pyi +2 -2
- metaflow-stubs/mf_extensions/outerbounds/plugins/card_utilities/injector.pyi +2 -2
- metaflow-stubs/mf_extensions/outerbounds/plugins/checkpoint_datastores/__init__.pyi +2 -2
- metaflow-stubs/mf_extensions/outerbounds/plugins/checkpoint_datastores/coreweave.pyi +2 -2
- metaflow-stubs/mf_extensions/outerbounds/plugins/checkpoint_datastores/nebius.pyi +2 -2
- metaflow-stubs/mf_extensions/outerbounds/plugins/fast_bakery/__init__.pyi +2 -2
- metaflow-stubs/mf_extensions/outerbounds/plugins/fast_bakery/baker.pyi +4 -4
- metaflow-stubs/mf_extensions/outerbounds/plugins/fast_bakery/docker_environment.pyi +2 -2
- metaflow-stubs/mf_extensions/outerbounds/plugins/fast_bakery/fast_bakery.pyi +2 -2
- metaflow-stubs/mf_extensions/outerbounds/plugins/kubernetes/__init__.pyi +2 -2
- metaflow-stubs/mf_extensions/outerbounds/plugins/kubernetes/pod_killer.pyi +2 -2
- metaflow-stubs/mf_extensions/outerbounds/plugins/ollama/__init__.pyi +2 -2
- metaflow-stubs/mf_extensions/outerbounds/plugins/ollama/constants.pyi +2 -2
- metaflow-stubs/mf_extensions/outerbounds/plugins/ollama/exceptions.pyi +2 -2
- metaflow-stubs/mf_extensions/outerbounds/plugins/ollama/ollama.pyi +2 -2
- metaflow-stubs/mf_extensions/outerbounds/plugins/ollama/status_card.pyi +2 -2
- metaflow-stubs/mf_extensions/outerbounds/plugins/snowflake/__init__.pyi +2 -2
- metaflow-stubs/mf_extensions/outerbounds/plugins/snowflake/snowflake.pyi +2 -2
- metaflow-stubs/mf_extensions/outerbounds/profilers/__init__.pyi +2 -2
- metaflow-stubs/mf_extensions/outerbounds/profilers/gpu.pyi +2 -2
- metaflow-stubs/mf_extensions/outerbounds/remote_config.pyi +2 -2
- metaflow-stubs/mf_extensions/outerbounds/toplevel/__init__.pyi +2 -2
- metaflow-stubs/mf_extensions/outerbounds/toplevel/global_aliases_for_metaflow_package.pyi +2 -2
- metaflow-stubs/multicore_utils.pyi +2 -2
- metaflow-stubs/ob_internal.pyi +2 -2
- metaflow-stubs/parameters.pyi +4 -4
- metaflow-stubs/plugins/__init__.pyi +13 -13
- metaflow-stubs/plugins/airflow/__init__.pyi +2 -2
- metaflow-stubs/plugins/airflow/airflow_utils.pyi +2 -2
- metaflow-stubs/plugins/airflow/exception.pyi +2 -2
- metaflow-stubs/plugins/airflow/sensors/__init__.pyi +2 -2
- metaflow-stubs/plugins/airflow/sensors/base_sensor.pyi +2 -2
- metaflow-stubs/plugins/airflow/sensors/external_task_sensor.pyi +2 -2
- metaflow-stubs/plugins/airflow/sensors/s3_sensor.pyi +2 -2
- metaflow-stubs/plugins/argo/__init__.pyi +2 -2
- metaflow-stubs/plugins/argo/argo_client.pyi +2 -2
- metaflow-stubs/plugins/argo/argo_events.pyi +2 -2
- metaflow-stubs/plugins/argo/argo_workflows.pyi +3 -3
- metaflow-stubs/plugins/argo/argo_workflows_decorator.pyi +3 -3
- metaflow-stubs/plugins/argo/argo_workflows_deployer.pyi +3 -3
- metaflow-stubs/plugins/argo/argo_workflows_deployer_objects.pyi +3 -3
- metaflow-stubs/plugins/aws/__init__.pyi +2 -2
- metaflow-stubs/plugins/aws/aws_client.pyi +2 -2
- metaflow-stubs/plugins/aws/aws_utils.pyi +2 -2
- metaflow-stubs/plugins/aws/batch/__init__.pyi +2 -2
- metaflow-stubs/plugins/aws/batch/batch.pyi +2 -2
- metaflow-stubs/plugins/aws/batch/batch_client.pyi +2 -2
- metaflow-stubs/plugins/aws/batch/batch_decorator.pyi +2 -2
- metaflow-stubs/plugins/aws/secrets_manager/__init__.pyi +2 -2
- metaflow-stubs/plugins/aws/secrets_manager/aws_secrets_manager_secrets_provider.pyi +2 -2
- metaflow-stubs/plugins/aws/step_functions/__init__.pyi +2 -2
- metaflow-stubs/plugins/aws/step_functions/event_bridge_client.pyi +2 -2
- metaflow-stubs/plugins/aws/step_functions/schedule_decorator.pyi +2 -2
- metaflow-stubs/plugins/aws/step_functions/step_functions.pyi +2 -2
- metaflow-stubs/plugins/aws/step_functions/step_functions_client.pyi +2 -2
- metaflow-stubs/plugins/aws/step_functions/step_functions_deployer.pyi +4 -4
- metaflow-stubs/plugins/aws/step_functions/step_functions_deployer_objects.pyi +4 -4
- metaflow-stubs/plugins/azure/__init__.pyi +2 -2
- metaflow-stubs/plugins/azure/azure_credential.pyi +2 -2
- metaflow-stubs/plugins/azure/azure_exceptions.pyi +2 -2
- metaflow-stubs/plugins/azure/azure_secret_manager_secrets_provider.pyi +2 -2
- metaflow-stubs/plugins/azure/azure_utils.pyi +2 -2
- metaflow-stubs/plugins/azure/blob_service_client_factory.pyi +2 -2
- metaflow-stubs/plugins/azure/includefile_support.pyi +2 -2
- metaflow-stubs/plugins/cards/__init__.pyi +6 -6
- metaflow-stubs/plugins/cards/card_client.pyi +2 -2
- metaflow-stubs/plugins/cards/card_creator.pyi +2 -2
- metaflow-stubs/plugins/cards/card_datastore.pyi +2 -4
- metaflow-stubs/plugins/cards/card_decorator.pyi +2 -2
- metaflow-stubs/plugins/cards/card_modules/__init__.pyi +2 -2
- metaflow-stubs/plugins/cards/card_modules/basic.pyi +2 -2
- metaflow-stubs/plugins/cards/card_modules/card.pyi +2 -2
- metaflow-stubs/plugins/cards/card_modules/components.pyi +4 -4
- metaflow-stubs/plugins/cards/card_modules/convert_to_native_type.pyi +2 -2
- metaflow-stubs/plugins/cards/card_modules/renderer_tools.pyi +2 -2
- metaflow-stubs/plugins/cards/card_modules/test_cards.pyi +2 -2
- metaflow-stubs/plugins/cards/card_resolver.pyi +2 -2
- metaflow-stubs/plugins/cards/component_serializer.pyi +2 -2
- metaflow-stubs/plugins/cards/exception.pyi +2 -2
- metaflow-stubs/plugins/catch_decorator.pyi +2 -2
- metaflow-stubs/plugins/datatools/__init__.pyi +2 -2
- metaflow-stubs/plugins/datatools/local.pyi +2 -2
- metaflow-stubs/plugins/datatools/s3/__init__.pyi +2 -2
- metaflow-stubs/plugins/datatools/s3/s3.pyi +3 -3
- metaflow-stubs/plugins/datatools/s3/s3tail.pyi +2 -2
- metaflow-stubs/plugins/datatools/s3/s3util.pyi +2 -2
- metaflow-stubs/plugins/debug_logger.pyi +2 -2
- metaflow-stubs/plugins/debug_monitor.pyi +2 -2
- metaflow-stubs/plugins/environment_decorator.pyi +2 -2
- metaflow-stubs/plugins/events_decorator.pyi +2 -2
- metaflow-stubs/plugins/frameworks/__init__.pyi +2 -2
- metaflow-stubs/plugins/frameworks/pytorch.pyi +2 -2
- metaflow-stubs/plugins/gcp/__init__.pyi +2 -2
- metaflow-stubs/plugins/gcp/gcp_secret_manager_secrets_provider.pyi +2 -2
- metaflow-stubs/plugins/gcp/gs_exceptions.pyi +2 -2
- metaflow-stubs/plugins/gcp/gs_storage_client_factory.pyi +2 -2
- metaflow-stubs/plugins/gcp/gs_utils.pyi +2 -2
- metaflow-stubs/plugins/gcp/includefile_support.pyi +2 -2
- metaflow-stubs/plugins/kubernetes/__init__.pyi +2 -2
- metaflow-stubs/plugins/kubernetes/kube_utils.pyi +3 -3
- metaflow-stubs/plugins/kubernetes/kubernetes.pyi +2 -2
- metaflow-stubs/plugins/kubernetes/kubernetes_client.pyi +2 -2
- metaflow-stubs/plugins/kubernetes/kubernetes_decorator.pyi +2 -2
- metaflow-stubs/plugins/kubernetes/kubernetes_jobsets.pyi +2 -2
- metaflow-stubs/plugins/kubernetes/spot_monitor_sidecar.pyi +2 -2
- metaflow-stubs/plugins/ollama/__init__.pyi +3 -3
- metaflow-stubs/plugins/parallel_decorator.pyi +2 -2
- metaflow-stubs/plugins/perimeters.pyi +2 -2
- metaflow-stubs/plugins/project_decorator.pyi +2 -2
- metaflow-stubs/plugins/pypi/__init__.pyi +3 -3
- metaflow-stubs/plugins/pypi/conda_decorator.pyi +2 -2
- metaflow-stubs/plugins/pypi/conda_environment.pyi +5 -5
- metaflow-stubs/plugins/pypi/parsers.pyi +2 -2
- metaflow-stubs/plugins/pypi/pypi_decorator.pyi +2 -2
- metaflow-stubs/plugins/pypi/pypi_environment.pyi +2 -2
- metaflow-stubs/plugins/pypi/utils.pyi +2 -2
- metaflow-stubs/plugins/resources_decorator.pyi +2 -2
- metaflow-stubs/plugins/retry_decorator.pyi +2 -2
- metaflow-stubs/plugins/secrets/__init__.pyi +2 -2
- metaflow-stubs/plugins/secrets/inline_secrets_provider.pyi +2 -2
- metaflow-stubs/plugins/secrets/secrets_decorator.pyi +2 -2
- metaflow-stubs/plugins/snowflake/__init__.pyi +2 -2
- metaflow-stubs/plugins/storage_executor.pyi +2 -2
- metaflow-stubs/plugins/test_unbounded_foreach_decorator.pyi +3 -3
- metaflow-stubs/plugins/timeout_decorator.pyi +2 -2
- metaflow-stubs/plugins/torchtune/__init__.pyi +2 -2
- metaflow-stubs/plugins/uv/__init__.pyi +2 -2
- metaflow-stubs/plugins/uv/uv_environment.pyi +2 -2
- metaflow-stubs/profilers/__init__.pyi +2 -2
- metaflow-stubs/pylint_wrapper.pyi +2 -2
- metaflow-stubs/runner/__init__.pyi +2 -2
- metaflow-stubs/runner/deployer.pyi +5 -5
- metaflow-stubs/runner/deployer_impl.pyi +2 -2
- metaflow-stubs/runner/metaflow_runner.pyi +4 -4
- metaflow-stubs/runner/nbdeploy.pyi +2 -2
- metaflow-stubs/runner/nbrun.pyi +2 -2
- metaflow-stubs/runner/subprocess_manager.pyi +2 -2
- metaflow-stubs/runner/utils.pyi +4 -4
- metaflow-stubs/system/__init__.pyi +2 -2
- metaflow-stubs/system/system_logger.pyi +3 -3
- metaflow-stubs/system/system_monitor.pyi +2 -2
- metaflow-stubs/tagging_util.pyi +2 -2
- metaflow-stubs/tuple_util.pyi +2 -2
- metaflow-stubs/user_configs/__init__.pyi +2 -2
- metaflow-stubs/user_configs/config_decorators.pyi +5 -5
- metaflow-stubs/user_configs/config_options.pyi +3 -3
- metaflow-stubs/user_configs/config_parameters.pyi +6 -6
- {ob_metaflow_stubs-6.0.3.182rc1.dist-info → ob_metaflow_stubs-6.0.3.183rc0.dist-info}/METADATA +1 -1
- ob_metaflow_stubs-6.0.3.183rc0.dist-info/RECORD +220 -0
- ob_metaflow_stubs-6.0.3.182rc1.dist-info/RECORD +0 -220
- {ob_metaflow_stubs-6.0.3.182rc1.dist-info → ob_metaflow_stubs-6.0.3.183rc0.dist-info}/WHEEL +0 -0
- {ob_metaflow_stubs-6.0.3.182rc1.dist-info → ob_metaflow_stubs-6.0.3.183rc0.dist-info}/top_level.txt +0 -0
metaflow-stubs/__init__.pyi
CHANGED
@@ -1,7 +1,7 @@
|
|
1
1
|
######################################################################################################
|
2
2
|
# Auto-generated Metaflow stub file #
|
3
|
-
# MF version: 2.15.
|
4
|
-
# Generated on 2025-06-
|
3
|
+
# MF version: 2.15.18.1+obcheckpoint(0.2.1);ob(v1) #
|
4
|
+
# Generated on 2025-06-20T20:53:54.419355 #
|
5
5
|
######################################################################################################
|
6
6
|
|
7
7
|
from __future__ import annotations
|
@@ -35,9 +35,9 @@ from .user_configs.config_parameters import ConfigValue as ConfigValue
|
|
35
35
|
from .user_configs.config_parameters import config_expr as config_expr
|
36
36
|
from .user_configs.config_decorators import CustomFlowDecorator as CustomFlowDecorator
|
37
37
|
from .user_configs.config_decorators import CustomStepDecorator as CustomStepDecorator
|
38
|
-
from . import metaflow_git as metaflow_git
|
39
38
|
from . import tuple_util as tuple_util
|
40
39
|
from . import cards as cards
|
40
|
+
from . import metaflow_git as metaflow_git
|
41
41
|
from . import events as events
|
42
42
|
from . import runner as runner
|
43
43
|
from . import plugins as plugins
|
@@ -156,53 +156,170 @@ def step(f: typing.Union[typing.Callable[[FlowSpecDerived], None], typing.Callab
|
|
156
156
|
...
|
157
157
|
|
158
158
|
@typing.overload
|
159
|
-
def
|
159
|
+
def fast_bakery_internal(f: typing.Callable[[FlowSpecDerived, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, StepFlag], None]:
|
160
160
|
"""
|
161
|
-
|
161
|
+
Internal decorator to support Fast bakery
|
162
|
+
"""
|
163
|
+
...
|
164
|
+
|
165
|
+
@typing.overload
|
166
|
+
def fast_bakery_internal(f: typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]:
|
167
|
+
...
|
168
|
+
|
169
|
+
def fast_bakery_internal(f: typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None], None] = None):
|
170
|
+
"""
|
171
|
+
Internal decorator to support Fast bakery
|
172
|
+
"""
|
173
|
+
...
|
174
|
+
|
175
|
+
@typing.overload
|
176
|
+
def secrets(*, sources: typing.List[typing.Union[str, typing.Dict[str, typing.Any]]] = []) -> typing.Callable[[typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]], typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]]:
|
177
|
+
"""
|
178
|
+
Specifies secrets to be retrieved and injected as environment variables prior to
|
179
|
+
the execution of a step.
|
162
180
|
|
163
|
-
|
164
|
-
|
165
|
-
|
166
|
-
|
181
|
+
|
182
|
+
Parameters
|
183
|
+
----------
|
184
|
+
sources : List[Union[str, Dict[str, Any]]], default: []
|
185
|
+
List of secret specs, defining how the secrets are to be retrieved
|
186
|
+
"""
|
187
|
+
...
|
188
|
+
|
189
|
+
@typing.overload
|
190
|
+
def secrets(f: typing.Callable[[FlowSpecDerived, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, StepFlag], None]:
|
191
|
+
...
|
192
|
+
|
193
|
+
@typing.overload
|
194
|
+
def secrets(f: typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]:
|
195
|
+
...
|
196
|
+
|
197
|
+
def secrets(f: typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None], None] = None, *, sources: typing.List[typing.Union[str, typing.Dict[str, typing.Any]]] = []):
|
198
|
+
"""
|
199
|
+
Specifies secrets to be retrieved and injected as environment variables prior to
|
200
|
+
the execution of a step.
|
167
201
|
|
168
202
|
|
169
203
|
Parameters
|
170
204
|
----------
|
171
|
-
|
172
|
-
|
173
|
-
If not specified, the exception is not stored.
|
174
|
-
print_exception : bool, default True
|
175
|
-
Determines whether or not the exception is printed to
|
176
|
-
stdout when caught.
|
205
|
+
sources : List[Union[str, Dict[str, Any]]], default: []
|
206
|
+
List of secret specs, defining how the secrets are to be retrieved
|
177
207
|
"""
|
178
208
|
...
|
179
209
|
|
180
210
|
@typing.overload
|
181
|
-
def
|
211
|
+
def model(*, load: typing.Union[typing.List[str], str, typing.List[typing.Tuple[str, typing.Optional[str]]]] = None, temp_dir_root: str = None) -> typing.Callable[[typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]], typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]]:
|
212
|
+
"""
|
213
|
+
Enables loading / saving of models within a step.
|
214
|
+
|
215
|
+
|
216
|
+
|
217
|
+
Parameters
|
218
|
+
----------
|
219
|
+
load : Union[List[str],str,List[Tuple[str,Union[str,None]]]], default: None
|
220
|
+
Artifact name/s referencing the models/checkpoints to load. Artifact names refer to the names of the instance variables set to `self`.
|
221
|
+
These artifact names give to `load` be reference objects or reference `key` string's from objects created by:
|
222
|
+
- `current.checkpoint`
|
223
|
+
- `current.model`
|
224
|
+
- `current.huggingface_hub`
|
225
|
+
|
226
|
+
If a list of tuples is provided, the first element is the artifact name and the second element is the path the artifact needs be unpacked on
|
227
|
+
the local filesystem. If the second element is None, the artifact will be unpacked in the current working directory.
|
228
|
+
If a string is provided, then the artifact corresponding to that name will be loaded in the current working directory.
|
229
|
+
|
230
|
+
temp_dir_root : str, default: None
|
231
|
+
The root directory under which `current.model.loaded` will store loaded models
|
232
|
+
"""
|
182
233
|
...
|
183
234
|
|
184
235
|
@typing.overload
|
185
|
-
def
|
236
|
+
def model(f: typing.Callable[[FlowSpecDerived, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, StepFlag], None]:
|
186
237
|
...
|
187
238
|
|
188
|
-
|
239
|
+
@typing.overload
|
240
|
+
def model(f: typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]:
|
241
|
+
...
|
242
|
+
|
243
|
+
def model(f: typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None], None] = None, *, load: typing.Union[typing.List[str], str, typing.List[typing.Tuple[str, typing.Optional[str]]]] = None, temp_dir_root: str = None):
|
189
244
|
"""
|
190
|
-
|
245
|
+
Enables loading / saving of models within a step.
|
191
246
|
|
192
|
-
The decorator will create an optional artifact, specified by `var`, which
|
193
|
-
contains the exception raised. You can use it to detect the presence
|
194
|
-
of errors, indicating that all happy-path artifacts produced by the step
|
195
|
-
are missing.
|
196
247
|
|
197
248
|
|
198
249
|
Parameters
|
199
250
|
----------
|
200
|
-
|
201
|
-
|
202
|
-
|
203
|
-
|
204
|
-
|
205
|
-
|
251
|
+
load : Union[List[str],str,List[Tuple[str,Union[str,None]]]], default: None
|
252
|
+
Artifact name/s referencing the models/checkpoints to load. Artifact names refer to the names of the instance variables set to `self`.
|
253
|
+
These artifact names give to `load` be reference objects or reference `key` string's from objects created by:
|
254
|
+
- `current.checkpoint`
|
255
|
+
- `current.model`
|
256
|
+
- `current.huggingface_hub`
|
257
|
+
|
258
|
+
If a list of tuples is provided, the first element is the artifact name and the second element is the path the artifact needs be unpacked on
|
259
|
+
the local filesystem. If the second element is None, the artifact will be unpacked in the current working directory.
|
260
|
+
If a string is provided, then the artifact corresponding to that name will be loaded in the current working directory.
|
261
|
+
|
262
|
+
temp_dir_root : str, default: None
|
263
|
+
The root directory under which `current.model.loaded` will store loaded models
|
264
|
+
"""
|
265
|
+
...
|
266
|
+
|
267
|
+
@typing.overload
|
268
|
+
def timeout(*, seconds: int = 0, minutes: int = 0, hours: int = 0) -> typing.Callable[[typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]], typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]]:
|
269
|
+
"""
|
270
|
+
Specifies a timeout for your step.
|
271
|
+
|
272
|
+
This decorator is useful if this step may hang indefinitely.
|
273
|
+
|
274
|
+
This can be used in conjunction with the `@retry` decorator as well as the `@catch` decorator.
|
275
|
+
A timeout is considered to be an exception thrown by the step. It will cause the step to be
|
276
|
+
retried if needed and the exception will be caught by the `@catch` decorator, if present.
|
277
|
+
|
278
|
+
Note that all the values specified in parameters are added together so if you specify
|
279
|
+
60 seconds and 1 hour, the decorator will have an effective timeout of 1 hour and 1 minute.
|
280
|
+
|
281
|
+
|
282
|
+
Parameters
|
283
|
+
----------
|
284
|
+
seconds : int, default 0
|
285
|
+
Number of seconds to wait prior to timing out.
|
286
|
+
minutes : int, default 0
|
287
|
+
Number of minutes to wait prior to timing out.
|
288
|
+
hours : int, default 0
|
289
|
+
Number of hours to wait prior to timing out.
|
290
|
+
"""
|
291
|
+
...
|
292
|
+
|
293
|
+
@typing.overload
|
294
|
+
def timeout(f: typing.Callable[[FlowSpecDerived, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, StepFlag], None]:
|
295
|
+
...
|
296
|
+
|
297
|
+
@typing.overload
|
298
|
+
def timeout(f: typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]:
|
299
|
+
...
|
300
|
+
|
301
|
+
def timeout(f: typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None], None] = None, *, seconds: int = 0, minutes: int = 0, hours: int = 0):
|
302
|
+
"""
|
303
|
+
Specifies a timeout for your step.
|
304
|
+
|
305
|
+
This decorator is useful if this step may hang indefinitely.
|
306
|
+
|
307
|
+
This can be used in conjunction with the `@retry` decorator as well as the `@catch` decorator.
|
308
|
+
A timeout is considered to be an exception thrown by the step. It will cause the step to be
|
309
|
+
retried if needed and the exception will be caught by the `@catch` decorator, if present.
|
310
|
+
|
311
|
+
Note that all the values specified in parameters are added together so if you specify
|
312
|
+
60 seconds and 1 hour, the decorator will have an effective timeout of 1 hour and 1 minute.
|
313
|
+
|
314
|
+
|
315
|
+
Parameters
|
316
|
+
----------
|
317
|
+
seconds : int, default 0
|
318
|
+
Number of seconds to wait prior to timing out.
|
319
|
+
minutes : int, default 0
|
320
|
+
Number of minutes to wait prior to timing out.
|
321
|
+
hours : int, default 0
|
322
|
+
Number of hours to wait prior to timing out.
|
206
323
|
"""
|
207
324
|
...
|
208
325
|
|
@@ -344,249 +461,85 @@ def card(f: typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typ
|
|
344
461
|
"""
|
345
462
|
...
|
346
463
|
|
347
|
-
|
348
|
-
def pypi(*, packages: typing.Dict[str, str] = {}, python: typing.Optional[str] = None) -> typing.Callable[[typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]], typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]]:
|
464
|
+
def app_deploy(*, app_port: int, app_name: str) -> typing.Callable[[typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]], typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]]:
|
349
465
|
"""
|
350
|
-
Specifies
|
351
|
-
|
352
|
-
Information in this decorator will augment any
|
353
|
-
attributes set in the `@pyi_base` flow-level decorator. Hence,
|
354
|
-
you can use `@pypi_base` to set packages required by all
|
355
|
-
steps and use `@pypi` to specify step-specific overrides.
|
466
|
+
Specifies that this step is used to deploy an instance of the app.
|
467
|
+
Requires that self.app_name, self.app_port, self.entrypoint and self.deployDir is set.
|
356
468
|
|
357
469
|
|
358
470
|
Parameters
|
359
471
|
----------
|
360
|
-
|
361
|
-
|
362
|
-
|
363
|
-
|
364
|
-
Version of Python to use, e.g. '3.7.4'. A default value of None implies
|
365
|
-
that the version used will correspond to the version of the Python interpreter used to start the run.
|
472
|
+
app_port : int
|
473
|
+
Number of GPUs to use.
|
474
|
+
app_name : str
|
475
|
+
Name of the app to deploy.
|
366
476
|
"""
|
367
477
|
...
|
368
478
|
|
369
479
|
@typing.overload
|
370
|
-
def
|
480
|
+
def parallel(f: typing.Callable[[FlowSpecDerived, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, StepFlag], None]:
|
481
|
+
"""
|
482
|
+
Decorator prototype for all step decorators. This function gets specialized
|
483
|
+
and imported for all decorators types by _import_plugin_decorators().
|
484
|
+
"""
|
371
485
|
...
|
372
486
|
|
373
487
|
@typing.overload
|
374
|
-
def
|
488
|
+
def parallel(f: typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]:
|
375
489
|
...
|
376
490
|
|
377
|
-
def
|
491
|
+
def parallel(f: typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None], None] = None):
|
378
492
|
"""
|
379
|
-
|
380
|
-
|
381
|
-
Information in this decorator will augment any
|
382
|
-
attributes set in the `@pyi_base` flow-level decorator. Hence,
|
383
|
-
you can use `@pypi_base` to set packages required by all
|
384
|
-
steps and use `@pypi` to specify step-specific overrides.
|
385
|
-
|
386
|
-
|
387
|
-
Parameters
|
388
|
-
----------
|
389
|
-
packages : Dict[str, str], default: {}
|
390
|
-
Packages to use for this step. The key is the name of the package
|
391
|
-
and the value is the version to use.
|
392
|
-
python : str, optional, default: None
|
393
|
-
Version of Python to use, e.g. '3.7.4'. A default value of None implies
|
394
|
-
that the version used will correspond to the version of the Python interpreter used to start the run.
|
493
|
+
Decorator prototype for all step decorators. This function gets specialized
|
494
|
+
and imported for all decorators types by _import_plugin_decorators().
|
395
495
|
"""
|
396
496
|
...
|
397
497
|
|
398
498
|
@typing.overload
|
399
|
-
def
|
499
|
+
def resources(*, cpu: int = 1, gpu: typing.Optional[int] = None, disk: typing.Optional[int] = None, memory: int = 4096, shared_memory: typing.Optional[int] = None) -> typing.Callable[[typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]], typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]]:
|
400
500
|
"""
|
401
|
-
|
501
|
+
Specifies the resources needed when executing this step.
|
402
502
|
|
503
|
+
Use `@resources` to specify the resource requirements
|
504
|
+
independently of the specific compute layer (`@batch`, `@kubernetes`).
|
505
|
+
|
506
|
+
You can choose the compute layer on the command line by executing e.g.
|
507
|
+
```
|
508
|
+
python myflow.py run --with batch
|
509
|
+
```
|
510
|
+
or
|
511
|
+
```
|
512
|
+
python myflow.py run --with kubernetes
|
513
|
+
```
|
514
|
+
which executes the flow on the desired system using the
|
515
|
+
requirements specified in `@resources`.
|
403
516
|
|
404
517
|
|
405
518
|
Parameters
|
406
519
|
----------
|
407
|
-
|
408
|
-
|
409
|
-
|
410
|
-
|
411
|
-
|
412
|
-
|
413
|
-
|
414
|
-
|
415
|
-
|
416
|
-
|
417
|
-
|
418
|
-
temp_dir_root : str, default: None
|
419
|
-
The root directory under which `current.model.loaded` will store loaded models
|
420
|
-
"""
|
421
|
-
...
|
422
|
-
|
423
|
-
@typing.overload
|
424
|
-
def model(f: typing.Callable[[FlowSpecDerived, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, StepFlag], None]:
|
425
|
-
...
|
426
|
-
|
427
|
-
@typing.overload
|
428
|
-
def model(f: typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]:
|
429
|
-
...
|
430
|
-
|
431
|
-
def model(f: typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None], None] = None, *, load: typing.Union[typing.List[str], str, typing.List[typing.Tuple[str, typing.Optional[str]]]] = None, temp_dir_root: str = None):
|
432
|
-
"""
|
433
|
-
Enables loading / saving of models within a step.
|
434
|
-
|
435
|
-
|
436
|
-
|
437
|
-
Parameters
|
438
|
-
----------
|
439
|
-
load : Union[List[str],str,List[Tuple[str,Union[str,None]]]], default: None
|
440
|
-
Artifact name/s referencing the models/checkpoints to load. Artifact names refer to the names of the instance variables set to `self`.
|
441
|
-
These artifact names give to `load` be reference objects or reference `key` string's from objects created by:
|
442
|
-
- `current.checkpoint`
|
443
|
-
- `current.model`
|
444
|
-
- `current.huggingface_hub`
|
445
|
-
|
446
|
-
If a list of tuples is provided, the first element is the artifact name and the second element is the path the artifact needs be unpacked on
|
447
|
-
the local filesystem. If the second element is None, the artifact will be unpacked in the current working directory.
|
448
|
-
If a string is provided, then the artifact corresponding to that name will be loaded in the current working directory.
|
449
|
-
|
450
|
-
temp_dir_root : str, default: None
|
451
|
-
The root directory under which `current.model.loaded` will store loaded models
|
452
|
-
"""
|
453
|
-
...
|
454
|
-
|
455
|
-
@typing.overload
|
456
|
-
def checkpoint(*, load_policy: str = 'fresh', temp_dir_root: str = None) -> typing.Callable[[typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]], typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]]:
|
457
|
-
"""
|
458
|
-
Enables checkpointing for a step.
|
459
|
-
|
460
|
-
|
461
|
-
|
462
|
-
Parameters
|
463
|
-
----------
|
464
|
-
load_policy : str, default: "fresh"
|
465
|
-
The policy for loading the checkpoint. The following policies are supported:
|
466
|
-
- "eager": Loads the the latest available checkpoint within the namespace.
|
467
|
-
With this mode, the latest checkpoint written by any previous task (can be even a different run) of the step
|
468
|
-
will be loaded at the start of the task.
|
469
|
-
- "none": Do not load any checkpoint
|
470
|
-
- "fresh": Loads the lastest checkpoint created within the running Task.
|
471
|
-
This mode helps loading checkpoints across various retry attempts of the same task.
|
472
|
-
With this mode, no checkpoint will be loaded at the start of a task but any checkpoints
|
473
|
-
created within the task will be loaded when the task is retries execution on failure.
|
474
|
-
|
475
|
-
temp_dir_root : str, default: None
|
476
|
-
The root directory under which `current.checkpoint.directory` will be created.
|
477
|
-
"""
|
478
|
-
...
|
479
|
-
|
480
|
-
@typing.overload
|
481
|
-
def checkpoint(f: typing.Callable[[FlowSpecDerived, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, StepFlag], None]:
|
482
|
-
...
|
483
|
-
|
484
|
-
@typing.overload
|
485
|
-
def checkpoint(f: typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]:
|
486
|
-
...
|
487
|
-
|
488
|
-
def checkpoint(f: typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None], None] = None, *, load_policy: str = 'fresh', temp_dir_root: str = None):
|
489
|
-
"""
|
490
|
-
Enables checkpointing for a step.
|
491
|
-
|
492
|
-
|
493
|
-
|
494
|
-
Parameters
|
495
|
-
----------
|
496
|
-
load_policy : str, default: "fresh"
|
497
|
-
The policy for loading the checkpoint. The following policies are supported:
|
498
|
-
- "eager": Loads the the latest available checkpoint within the namespace.
|
499
|
-
With this mode, the latest checkpoint written by any previous task (can be even a different run) of the step
|
500
|
-
will be loaded at the start of the task.
|
501
|
-
- "none": Do not load any checkpoint
|
502
|
-
- "fresh": Loads the lastest checkpoint created within the running Task.
|
503
|
-
This mode helps loading checkpoints across various retry attempts of the same task.
|
504
|
-
With this mode, no checkpoint will be loaded at the start of a task but any checkpoints
|
505
|
-
created within the task will be loaded when the task is retries execution on failure.
|
506
|
-
|
507
|
-
temp_dir_root : str, default: None
|
508
|
-
The root directory under which `current.checkpoint.directory` will be created.
|
509
|
-
"""
|
510
|
-
...
|
511
|
-
|
512
|
-
@typing.overload
|
513
|
-
def secrets(*, sources: typing.List[typing.Union[str, typing.Dict[str, typing.Any]]] = []) -> typing.Callable[[typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]], typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]]:
|
514
|
-
"""
|
515
|
-
Specifies secrets to be retrieved and injected as environment variables prior to
|
516
|
-
the execution of a step.
|
517
|
-
|
518
|
-
|
519
|
-
Parameters
|
520
|
-
----------
|
521
|
-
sources : List[Union[str, Dict[str, Any]]], default: []
|
522
|
-
List of secret specs, defining how the secrets are to be retrieved
|
520
|
+
cpu : int, default 1
|
521
|
+
Number of CPUs required for this step.
|
522
|
+
gpu : int, optional, default None
|
523
|
+
Number of GPUs required for this step.
|
524
|
+
disk : int, optional, default None
|
525
|
+
Disk size (in MB) required for this step. Only applies on Kubernetes.
|
526
|
+
memory : int, default 4096
|
527
|
+
Memory size (in MB) required for this step.
|
528
|
+
shared_memory : int, optional, default None
|
529
|
+
The value for the size (in MiB) of the /dev/shm volume for this step.
|
530
|
+
This parameter maps to the `--shm-size` option in Docker.
|
523
531
|
"""
|
524
532
|
...
|
525
533
|
|
526
534
|
@typing.overload
|
527
|
-
def
|
535
|
+
def resources(f: typing.Callable[[FlowSpecDerived, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, StepFlag], None]:
|
528
536
|
...
|
529
537
|
|
530
538
|
@typing.overload
|
531
|
-
def
|
532
|
-
...
|
533
|
-
|
534
|
-
def secrets(f: typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None], None] = None, *, sources: typing.List[typing.Union[str, typing.Dict[str, typing.Any]]] = []):
|
535
|
-
"""
|
536
|
-
Specifies secrets to be retrieved and injected as environment variables prior to
|
537
|
-
the execution of a step.
|
538
|
-
|
539
|
-
|
540
|
-
Parameters
|
541
|
-
----------
|
542
|
-
sources : List[Union[str, Dict[str, Any]]], default: []
|
543
|
-
List of secret specs, defining how the secrets are to be retrieved
|
544
|
-
"""
|
545
|
-
...
|
546
|
-
|
547
|
-
def nvidia(*, gpu: int, gpu_type: str, queue_timeout: int) -> typing.Callable[[typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]], typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]]:
|
548
|
-
"""
|
549
|
-
Specifies that this step should execute on DGX cloud.
|
550
|
-
|
551
|
-
|
552
|
-
Parameters
|
553
|
-
----------
|
554
|
-
gpu : int
|
555
|
-
Number of GPUs to use.
|
556
|
-
gpu_type : str
|
557
|
-
Type of Nvidia GPU to use.
|
558
|
-
queue_timeout : int
|
559
|
-
Time to keep the job in NVCF's queue.
|
560
|
-
"""
|
561
|
-
...
|
562
|
-
|
563
|
-
def huggingface_hub(*, temp_dir_root: typing.Optional[str] = None, load: typing.Union[typing.List[str], typing.List[typing.Tuple[typing.Dict, str]], typing.List[typing.Tuple[str, str]], typing.List[typing.Dict], None]) -> typing.Callable[[typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]], typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]]:
|
564
|
-
"""
|
565
|
-
Decorator that helps cache, version and store models/datasets from huggingface hub.
|
566
|
-
|
567
|
-
|
568
|
-
Parameters
|
569
|
-
----------
|
570
|
-
temp_dir_root : str, optional
|
571
|
-
The root directory that will hold the temporary directory where objects will be downloaded.
|
572
|
-
|
573
|
-
load: Union[List[str], List[Tuple[Dict, str]], List[Tuple[str, str]], List[Dict], None]
|
574
|
-
The list of repos (models/datasets) to load.
|
575
|
-
|
576
|
-
Loaded repos can be accessed via `current.huggingface_hub.loaded`. If load is set, then the following happens:
|
577
|
-
|
578
|
-
- If repo (model/dataset) is not found in the datastore:
|
579
|
-
- Downloads the repo from Hugging Face Hub to a temporary directory (or uses specified path) for local access
|
580
|
-
- Stores it in Metaflow's datastore (s3/gcs/azure etc.) with a unique name based on repo_type/repo_id
|
581
|
-
- All HF models loaded for a `@step` will be cached separately under flow/step/namespace.
|
582
|
-
|
583
|
-
- If repo is found in the datastore:
|
584
|
-
- Loads it directly from datastore to local path (can be temporary directory or specified path)
|
585
|
-
"""
|
539
|
+
def resources(f: typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]:
|
586
540
|
...
|
587
541
|
|
588
|
-
|
589
|
-
def resources(*, cpu: int = 1, gpu: typing.Optional[int] = None, disk: typing.Optional[int] = None, memory: int = 4096, shared_memory: typing.Optional[int] = None) -> typing.Callable[[typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]], typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]]:
|
542
|
+
def resources(f: typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None], None] = None, *, cpu: int = 1, gpu: typing.Optional[int] = None, disk: typing.Optional[int] = None, memory: int = 4096, shared_memory: typing.Optional[int] = None):
|
590
543
|
"""
|
591
544
|
Specifies the resources needed when executing this step.
|
592
545
|
|
@@ -622,45 +575,61 @@ def resources(*, cpu: int = 1, gpu: typing.Optional[int] = None, disk: typing.Op
|
|
622
575
|
...
|
623
576
|
|
624
577
|
@typing.overload
|
625
|
-
def
|
578
|
+
def conda(*, packages: typing.Dict[str, str] = {}, libraries: typing.Dict[str, str] = {}, python: typing.Optional[str] = None, disabled: bool = False) -> typing.Callable[[typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]], typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]]:
|
579
|
+
"""
|
580
|
+
Specifies the Conda environment for the step.
|
581
|
+
|
582
|
+
Information in this decorator will augment any
|
583
|
+
attributes set in the `@conda_base` flow-level decorator. Hence,
|
584
|
+
you can use `@conda_base` to set packages required by all
|
585
|
+
steps and use `@conda` to specify step-specific overrides.
|
586
|
+
|
587
|
+
|
588
|
+
Parameters
|
589
|
+
----------
|
590
|
+
packages : Dict[str, str], default {}
|
591
|
+
Packages to use for this step. The key is the name of the package
|
592
|
+
and the value is the version to use.
|
593
|
+
libraries : Dict[str, str], default {}
|
594
|
+
Supported for backward compatibility. When used with packages, packages will take precedence.
|
595
|
+
python : str, optional, default None
|
596
|
+
Version of Python to use, e.g. '3.7.4'. A default value of None implies
|
597
|
+
that the version used will correspond to the version of the Python interpreter used to start the run.
|
598
|
+
disabled : bool, default False
|
599
|
+
If set to True, disables @conda.
|
600
|
+
"""
|
626
601
|
...
|
627
602
|
|
628
603
|
@typing.overload
|
629
|
-
def
|
604
|
+
def conda(f: typing.Callable[[FlowSpecDerived, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, StepFlag], None]:
|
630
605
|
...
|
631
606
|
|
632
|
-
|
607
|
+
@typing.overload
|
608
|
+
def conda(f: typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]:
|
609
|
+
...
|
610
|
+
|
611
|
+
def conda(f: typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None], None] = None, *, packages: typing.Dict[str, str] = {}, libraries: typing.Dict[str, str] = {}, python: typing.Optional[str] = None, disabled: bool = False):
|
633
612
|
"""
|
634
|
-
Specifies the
|
635
|
-
|
636
|
-
Use `@resources` to specify the resource requirements
|
637
|
-
independently of the specific compute layer (`@batch`, `@kubernetes`).
|
613
|
+
Specifies the Conda environment for the step.
|
638
614
|
|
639
|
-
|
640
|
-
|
641
|
-
|
642
|
-
|
643
|
-
or
|
644
|
-
```
|
645
|
-
python myflow.py run --with kubernetes
|
646
|
-
```
|
647
|
-
which executes the flow on the desired system using the
|
648
|
-
requirements specified in `@resources`.
|
615
|
+
Information in this decorator will augment any
|
616
|
+
attributes set in the `@conda_base` flow-level decorator. Hence,
|
617
|
+
you can use `@conda_base` to set packages required by all
|
618
|
+
steps and use `@conda` to specify step-specific overrides.
|
649
619
|
|
650
620
|
|
651
621
|
Parameters
|
652
622
|
----------
|
653
|
-
|
654
|
-
|
655
|
-
|
656
|
-
|
657
|
-
|
658
|
-
|
659
|
-
|
660
|
-
|
661
|
-
|
662
|
-
|
663
|
-
This parameter maps to the `--shm-size` option in Docker.
|
623
|
+
packages : Dict[str, str], default {}
|
624
|
+
Packages to use for this step. The key is the name of the package
|
625
|
+
and the value is the version to use.
|
626
|
+
libraries : Dict[str, str], default {}
|
627
|
+
Supported for backward compatibility. When used with packages, packages will take precedence.
|
628
|
+
python : str, optional, default None
|
629
|
+
Version of Python to use, e.g. '3.7.4'. A default value of None implies
|
630
|
+
that the version used will correspond to the version of the Python interpreter used to start the run.
|
631
|
+
disabled : bool, default False
|
632
|
+
If set to True, disables @conda.
|
664
633
|
"""
|
665
634
|
...
|
666
635
|
|
@@ -719,231 +688,117 @@ def retry(f: typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], ty
|
|
719
688
|
"""
|
720
689
|
...
|
721
690
|
|
722
|
-
def
|
691
|
+
def vllm(*, model: str, backend: str, debug: bool, kwargs: typing.Any) -> typing.Callable[[typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]], typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]]:
|
723
692
|
"""
|
724
|
-
This decorator is used to run
|
693
|
+
This decorator is used to run vllm APIs as Metaflow task sidecars.
|
725
694
|
|
726
695
|
User code call
|
727
696
|
--------------
|
728
|
-
@
|
729
|
-
|
697
|
+
@vllm(
|
698
|
+
model="...",
|
730
699
|
...
|
731
700
|
)
|
732
701
|
|
733
702
|
Valid backend options
|
734
703
|
---------------------
|
735
704
|
- 'local': Run as a separate process on the local task machine.
|
736
|
-
- (TODO) 'managed': Outerbounds hosts and selects compute provider.
|
737
|
-
- (TODO) 'remote': Spin up separate instance to serve Ollama models.
|
738
705
|
|
739
706
|
Valid model options
|
740
707
|
-------------------
|
741
|
-
Any model
|
708
|
+
Any HuggingFace model identifier, e.g. 'meta-llama/Llama-3.2-1B'
|
709
|
+
|
710
|
+
NOTE: vLLM's OpenAI-compatible server serves ONE model per server instance.
|
711
|
+
If you need multiple models, you must create multiple @vllm decorators.
|
742
712
|
|
743
713
|
|
744
714
|
Parameters
|
745
715
|
----------
|
746
|
-
|
747
|
-
|
716
|
+
model: str
|
717
|
+
HuggingFace model identifier to be served by vLLM.
|
748
718
|
backend: str
|
749
|
-
Determines where and how to run the
|
750
|
-
force_pull: bool
|
751
|
-
Whether to run `ollama pull` no matter what, or first check the remote cache in Metaflow datastore for this model key.
|
752
|
-
cache_update_policy: str
|
753
|
-
Cache update policy: "auto", "force", or "never".
|
754
|
-
force_cache_update: bool
|
755
|
-
Simple override for "force" cache update policy.
|
719
|
+
Determines where and how to run the vLLM process.
|
756
720
|
debug: bool
|
757
721
|
Whether to turn on verbose debugging logs.
|
758
|
-
|
759
|
-
|
760
|
-
|
761
|
-
|
762
|
-
"""
|
763
|
-
...
|
764
|
-
|
765
|
-
@typing.overload
|
766
|
-
def timeout(*, seconds: int = 0, minutes: int = 0, hours: int = 0) -> typing.Callable[[typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]], typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]]:
|
767
|
-
"""
|
768
|
-
Specifies a timeout for your step.
|
769
|
-
|
770
|
-
This decorator is useful if this step may hang indefinitely.
|
771
|
-
|
772
|
-
This can be used in conjunction with the `@retry` decorator as well as the `@catch` decorator.
|
773
|
-
A timeout is considered to be an exception thrown by the step. It will cause the step to be
|
774
|
-
retried if needed and the exception will be caught by the `@catch` decorator, if present.
|
775
|
-
|
776
|
-
Note that all the values specified in parameters are added together so if you specify
|
777
|
-
60 seconds and 1 hour, the decorator will have an effective timeout of 1 hour and 1 minute.
|
778
|
-
|
779
|
-
|
780
|
-
Parameters
|
781
|
-
----------
|
782
|
-
seconds : int, default 0
|
783
|
-
Number of seconds to wait prior to timing out.
|
784
|
-
minutes : int, default 0
|
785
|
-
Number of minutes to wait prior to timing out.
|
786
|
-
hours : int, default 0
|
787
|
-
Number of hours to wait prior to timing out.
|
788
|
-
"""
|
789
|
-
...
|
790
|
-
|
791
|
-
@typing.overload
|
792
|
-
def timeout(f: typing.Callable[[FlowSpecDerived, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, StepFlag], None]:
|
793
|
-
...
|
794
|
-
|
795
|
-
@typing.overload
|
796
|
-
def timeout(f: typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]:
|
797
|
-
...
|
798
|
-
|
799
|
-
def timeout(f: typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None], None] = None, *, seconds: int = 0, minutes: int = 0, hours: int = 0):
|
800
|
-
"""
|
801
|
-
Specifies a timeout for your step.
|
802
|
-
|
803
|
-
This decorator is useful if this step may hang indefinitely.
|
804
|
-
|
805
|
-
This can be used in conjunction with the `@retry` decorator as well as the `@catch` decorator.
|
806
|
-
A timeout is considered to be an exception thrown by the step. It will cause the step to be
|
807
|
-
retried if needed and the exception will be caught by the `@catch` decorator, if present.
|
808
|
-
|
809
|
-
Note that all the values specified in parameters are added together so if you specify
|
810
|
-
60 seconds and 1 hour, the decorator will have an effective timeout of 1 hour and 1 minute.
|
811
|
-
|
812
|
-
|
813
|
-
Parameters
|
814
|
-
----------
|
815
|
-
seconds : int, default 0
|
816
|
-
Number of seconds to wait prior to timing out.
|
817
|
-
minutes : int, default 0
|
818
|
-
Number of minutes to wait prior to timing out.
|
819
|
-
hours : int, default 0
|
820
|
-
Number of hours to wait prior to timing out.
|
722
|
+
kwargs : Any
|
723
|
+
Any other keyword arguments are passed directly to the vLLM engine.
|
724
|
+
This allows for flexible configuration of vLLM server settings.
|
725
|
+
For example, `tensor_parallel_size=2`.
|
821
726
|
"""
|
822
727
|
...
|
823
728
|
|
824
729
|
@typing.overload
|
825
|
-
def
|
730
|
+
def pypi(*, packages: typing.Dict[str, str] = {}, python: typing.Optional[str] = None) -> typing.Callable[[typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]], typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]]:
|
826
731
|
"""
|
827
|
-
Specifies the
|
732
|
+
Specifies the PyPI packages for the step.
|
828
733
|
|
829
734
|
Information in this decorator will augment any
|
830
|
-
attributes set in the `@
|
831
|
-
you can use `@
|
832
|
-
steps and use `@
|
735
|
+
attributes set in the `@pyi_base` flow-level decorator. Hence,
|
736
|
+
you can use `@pypi_base` to set packages required by all
|
737
|
+
steps and use `@pypi` to specify step-specific overrides.
|
833
738
|
|
834
739
|
|
835
740
|
Parameters
|
836
741
|
----------
|
837
|
-
packages : Dict[str, str], default {}
|
742
|
+
packages : Dict[str, str], default: {}
|
838
743
|
Packages to use for this step. The key is the name of the package
|
839
744
|
and the value is the version to use.
|
840
|
-
|
841
|
-
Supported for backward compatibility. When used with packages, packages will take precedence.
|
842
|
-
python : str, optional, default None
|
745
|
+
python : str, optional, default: None
|
843
746
|
Version of Python to use, e.g. '3.7.4'. A default value of None implies
|
844
747
|
that the version used will correspond to the version of the Python interpreter used to start the run.
|
845
|
-
disabled : bool, default False
|
846
|
-
If set to True, disables @conda.
|
847
748
|
"""
|
848
749
|
...
|
849
750
|
|
850
751
|
@typing.overload
|
851
|
-
def
|
752
|
+
def pypi(f: typing.Callable[[FlowSpecDerived, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, StepFlag], None]:
|
852
753
|
...
|
853
754
|
|
854
755
|
@typing.overload
|
855
|
-
def
|
756
|
+
def pypi(f: typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]:
|
856
757
|
...
|
857
758
|
|
858
|
-
def
|
759
|
+
def pypi(f: typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None], None] = None, *, packages: typing.Dict[str, str] = {}, python: typing.Optional[str] = None):
|
859
760
|
"""
|
860
|
-
Specifies the
|
761
|
+
Specifies the PyPI packages for the step.
|
861
762
|
|
862
763
|
Information in this decorator will augment any
|
863
|
-
attributes set in the `@
|
864
|
-
you can use `@
|
865
|
-
steps and use `@
|
764
|
+
attributes set in the `@pyi_base` flow-level decorator. Hence,
|
765
|
+
you can use `@pypi_base` to set packages required by all
|
766
|
+
steps and use `@pypi` to specify step-specific overrides.
|
866
767
|
|
867
768
|
|
868
769
|
Parameters
|
869
770
|
----------
|
870
|
-
packages : Dict[str, str], default {}
|
771
|
+
packages : Dict[str, str], default: {}
|
871
772
|
Packages to use for this step. The key is the name of the package
|
872
773
|
and the value is the version to use.
|
873
|
-
|
874
|
-
Supported for backward compatibility. When used with packages, packages will take precedence.
|
875
|
-
python : str, optional, default None
|
774
|
+
python : str, optional, default: None
|
876
775
|
Version of Python to use, e.g. '3.7.4'. A default value of None implies
|
877
776
|
that the version used will correspond to the version of the Python interpreter used to start the run.
|
878
|
-
disabled : bool, default False
|
879
|
-
If set to True, disables @conda.
|
880
|
-
"""
|
881
|
-
...
|
882
|
-
|
883
|
-
@typing.overload
|
884
|
-
def parallel(f: typing.Callable[[FlowSpecDerived, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, StepFlag], None]:
|
885
|
-
"""
|
886
|
-
Decorator prototype for all step decorators. This function gets specialized
|
887
|
-
and imported for all decorators types by _import_plugin_decorators().
|
888
|
-
"""
|
889
|
-
...
|
890
|
-
|
891
|
-
@typing.overload
|
892
|
-
def parallel(f: typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]:
|
893
|
-
...
|
894
|
-
|
895
|
-
def parallel(f: typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None], None] = None):
|
896
|
-
"""
|
897
|
-
Decorator prototype for all step decorators. This function gets specialized
|
898
|
-
and imported for all decorators types by _import_plugin_decorators().
|
899
777
|
"""
|
900
778
|
...
|
901
779
|
|
902
|
-
|
903
|
-
def environment(*, vars: typing.Dict[str, str] = {}) -> typing.Callable[[typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]], typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]]:
|
780
|
+
def huggingface_hub(*, temp_dir_root: typing.Optional[str] = None, load: typing.Union[typing.List[str], typing.List[typing.Tuple[typing.Dict, str]], typing.List[typing.Tuple[str, str]], typing.List[typing.Dict], None]) -> typing.Callable[[typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]], typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]]:
|
904
781
|
"""
|
905
|
-
|
782
|
+
Decorator that helps cache, version and store models/datasets from huggingface hub.
|
906
783
|
|
907
784
|
|
908
785
|
Parameters
|
909
786
|
----------
|
910
|
-
|
911
|
-
|
912
|
-
"""
|
913
|
-
...
|
914
|
-
|
915
|
-
@typing.overload
|
916
|
-
def environment(f: typing.Callable[[FlowSpecDerived, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, StepFlag], None]:
|
917
|
-
...
|
918
|
-
|
919
|
-
@typing.overload
|
920
|
-
def environment(f: typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]:
|
921
|
-
...
|
922
|
-
|
923
|
-
def environment(f: typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None], None] = None, *, vars: typing.Dict[str, str] = {}):
|
924
|
-
"""
|
925
|
-
Specifies environment variables to be set prior to the execution of a step.
|
787
|
+
temp_dir_root : str, optional
|
788
|
+
The root directory that will hold the temporary directory where objects will be downloaded.
|
926
789
|
|
790
|
+
load: Union[List[str], List[Tuple[Dict, str]], List[Tuple[str, str]], List[Dict], None]
|
791
|
+
The list of repos (models/datasets) to load.
|
927
792
|
|
928
|
-
|
929
|
-
----------
|
930
|
-
vars : Dict[str, str], default {}
|
931
|
-
Dictionary of environment variables to set.
|
932
|
-
"""
|
933
|
-
...
|
934
|
-
|
935
|
-
def app_deploy(*, app_port: int, app_name: str) -> typing.Callable[[typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]], typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]]:
|
936
|
-
"""
|
937
|
-
Specifies that this step is used to deploy an instance of the app.
|
938
|
-
Requires that self.app_name, self.app_port, self.entrypoint and self.deployDir is set.
|
793
|
+
Loaded repos can be accessed via `current.huggingface_hub.loaded`. If load is set, then the following happens:
|
939
794
|
|
795
|
+
- If repo (model/dataset) is not found in the datastore:
|
796
|
+
- Downloads the repo from Hugging Face Hub to a temporary directory (or uses specified path) for local access
|
797
|
+
- Stores it in Metaflow's datastore (s3/gcs/azure etc.) with a unique name based on repo_type/repo_id
|
798
|
+
- All HF models loaded for a `@step` will be cached separately under flow/step/namespace.
|
940
799
|
|
941
|
-
|
942
|
-
|
943
|
-
app_port : int
|
944
|
-
Number of GPUs to use.
|
945
|
-
app_name : str
|
946
|
-
Name of the app to deploy.
|
800
|
+
- If repo is found in the datastore:
|
801
|
+
- Loads it directly from datastore to local path (can be temporary directory or specified path)
|
947
802
|
"""
|
948
803
|
...
|
949
804
|
|
@@ -961,291 +816,203 @@ def nvct(*, gpu: int, gpu_type: str) -> typing.Callable[[typing.Union[typing.Cal
|
|
961
816
|
"""
|
962
817
|
...
|
963
818
|
|
964
|
-
def
|
819
|
+
def ollama(*, models: list, backend: str, force_pull: bool, cache_update_policy: str, force_cache_update: bool, debug: bool, circuit_breaker_config: dict, timeout_config: dict) -> typing.Callable[[typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]], typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]]:
|
965
820
|
"""
|
966
|
-
This decorator is used to run
|
821
|
+
This decorator is used to run Ollama APIs as Metaflow task sidecars.
|
967
822
|
|
968
823
|
User code call
|
969
824
|
--------------
|
970
|
-
@
|
971
|
-
|
825
|
+
@ollama(
|
826
|
+
models=[...],
|
972
827
|
...
|
973
828
|
)
|
974
829
|
|
975
830
|
Valid backend options
|
976
831
|
---------------------
|
977
832
|
- 'local': Run as a separate process on the local task machine.
|
833
|
+
- (TODO) 'managed': Outerbounds hosts and selects compute provider.
|
834
|
+
- (TODO) 'remote': Spin up separate instance to serve Ollama models.
|
978
835
|
|
979
836
|
Valid model options
|
980
837
|
-------------------
|
981
|
-
Any
|
982
|
-
|
983
|
-
NOTE: vLLM's OpenAI-compatible server serves ONE model per server instance.
|
984
|
-
If you need multiple models, you must create multiple @vllm decorators.
|
838
|
+
Any model here https://ollama.com/search, e.g. 'llama3.2', 'llama3.3'
|
985
839
|
|
986
840
|
|
987
841
|
Parameters
|
988
842
|
----------
|
989
|
-
|
990
|
-
|
843
|
+
models: list[str]
|
844
|
+
List of Ollama containers running models in sidecars.
|
991
845
|
backend: str
|
992
|
-
Determines where and how to run the
|
846
|
+
Determines where and how to run the Ollama process.
|
847
|
+
force_pull: bool
|
848
|
+
Whether to run `ollama pull` no matter what, or first check the remote cache in Metaflow datastore for this model key.
|
849
|
+
cache_update_policy: str
|
850
|
+
Cache update policy: "auto", "force", or "never".
|
851
|
+
force_cache_update: bool
|
852
|
+
Simple override for "force" cache update policy.
|
993
853
|
debug: bool
|
994
854
|
Whether to turn on verbose debugging logs.
|
995
|
-
|
996
|
-
|
997
|
-
|
998
|
-
|
855
|
+
circuit_breaker_config: dict
|
856
|
+
Configuration for circuit breaker protection. Keys: failure_threshold, recovery_timeout, reset_timeout.
|
857
|
+
timeout_config: dict
|
858
|
+
Configuration for various operation timeouts. Keys: pull, stop, health_check, install, server_startup.
|
999
859
|
"""
|
1000
860
|
...
|
1001
861
|
|
1002
|
-
|
1003
|
-
def fast_bakery_internal(f: typing.Callable[[FlowSpecDerived, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, StepFlag], None]:
|
862
|
+
def nvidia(*, gpu: int, gpu_type: str, queue_timeout: int) -> typing.Callable[[typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]], typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]]:
|
1004
863
|
"""
|
1005
|
-
|
864
|
+
Specifies that this step should execute on DGX cloud.
|
865
|
+
|
866
|
+
|
867
|
+
Parameters
|
868
|
+
----------
|
869
|
+
gpu : int
|
870
|
+
Number of GPUs to use.
|
871
|
+
gpu_type : str
|
872
|
+
Type of Nvidia GPU to use.
|
873
|
+
queue_timeout : int
|
874
|
+
Time to keep the job in NVCF's queue.
|
1006
875
|
"""
|
1007
876
|
...
|
1008
877
|
|
1009
878
|
@typing.overload
|
1010
|
-
def
|
1011
|
-
...
|
1012
|
-
|
1013
|
-
def fast_bakery_internal(f: typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None], None] = None):
|
1014
|
-
"""
|
1015
|
-
Internal decorator to support Fast bakery
|
1016
|
-
"""
|
1017
|
-
...
|
1018
|
-
|
1019
|
-
def with_artifact_store(f: typing.Optional[typing.Type[FlowSpecDerived]] = None):
|
879
|
+
def checkpoint(*, load_policy: str = 'fresh', temp_dir_root: str = None) -> typing.Callable[[typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]], typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]]:
|
1020
880
|
"""
|
1021
|
-
|
1022
|
-
`@checkpoint`/`@model`/`@huggingface_hub` decorators.
|
881
|
+
Enables checkpointing for a step.
|
1023
882
|
|
1024
|
-
This decorator is useful when users wish to save data to a different datastore
|
1025
|
-
than what is configured in Metaflow. This can be for variety of reasons:
|
1026
883
|
|
1027
|
-
1. Data security: The objects needs to be stored in a bucket (object storage) that is not accessible by other flows.
|
1028
|
-
2. Data Locality: The location where the task is executing is not located in the same region as the datastore.
|
1029
|
-
- Example: Metaflow datastore lives in US East, but the task is executing in Finland datacenters.
|
1030
|
-
3. Data Lifecycle Policies: The objects need to be archived / managed separately from the Metaflow managed objects.
|
1031
|
-
- Example: Flow is training very large models that need to be stored separately and will be deleted more aggressively than the Metaflow managed objects.
|
1032
884
|
|
1033
|
-
|
885
|
+
Parameters
|
1034
886
|
----------
|
887
|
+
load_policy : str, default: "fresh"
|
888
|
+
The policy for loading the checkpoint. The following policies are supported:
|
889
|
+
- "eager": Loads the the latest available checkpoint within the namespace.
|
890
|
+
With this mode, the latest checkpoint written by any previous task (can be even a different run) of the step
|
891
|
+
will be loaded at the start of the task.
|
892
|
+
- "none": Do not load any checkpoint
|
893
|
+
- "fresh": Loads the lastest checkpoint created within the running Task.
|
894
|
+
This mode helps loading checkpoints across various retry attempts of the same task.
|
895
|
+
With this mode, no checkpoint will be loaded at the start of a task but any checkpoints
|
896
|
+
created within the task will be loaded when the task is retries execution on failure.
|
1035
897
|
|
1036
|
-
|
1037
|
-
|
1038
|
-
|
1039
|
-
|
1040
|
-
|
1041
|
-
|
1042
|
-
|
1043
|
-
|
1044
|
-
|
1045
|
-
|
1046
|
-
|
1047
|
-
|
1048
|
-
|
1049
|
-
|
1050
|
-
|
1051
|
-
|
1052
|
-
f.write("Hello, World!")
|
1053
|
-
self.external_bucket_checkpoint = current.checkpoint.save("my_file.txt")
|
1054
|
-
self.next(self.end)
|
1055
|
-
|
1056
|
-
```
|
1057
|
-
|
1058
|
-
- Using credentials to access the s3-compatible datastore.
|
1059
|
-
|
1060
|
-
```python
|
1061
|
-
@with_artifact_store(
|
1062
|
-
type="s3",
|
1063
|
-
config=lambda: {
|
1064
|
-
"root": "s3://my-bucket-foo/path/to/root",
|
1065
|
-
"client_params": {
|
1066
|
-
"aws_access_key_id": os.environ.get("MY_CUSTOM_ACCESS_KEY"),
|
1067
|
-
"aws_secret_access_key": os.environ.get("MY_CUSTOM_SECRET_KEY"),
|
1068
|
-
},
|
1069
|
-
},
|
1070
|
-
)
|
1071
|
-
class MyFlow(FlowSpec):
|
1072
|
-
|
1073
|
-
@checkpoint
|
1074
|
-
@step
|
1075
|
-
def start(self):
|
1076
|
-
with open("my_file.txt", "w") as f:
|
1077
|
-
f.write("Hello, World!")
|
1078
|
-
self.external_bucket_checkpoint = current.checkpoint.save("my_file.txt")
|
1079
|
-
self.next(self.end)
|
1080
|
-
|
1081
|
-
```
|
898
|
+
temp_dir_root : str, default: None
|
899
|
+
The root directory under which `current.checkpoint.directory` will be created.
|
900
|
+
"""
|
901
|
+
...
|
902
|
+
|
903
|
+
@typing.overload
|
904
|
+
def checkpoint(f: typing.Callable[[FlowSpecDerived, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, StepFlag], None]:
|
905
|
+
...
|
906
|
+
|
907
|
+
@typing.overload
|
908
|
+
def checkpoint(f: typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]:
|
909
|
+
...
|
910
|
+
|
911
|
+
def checkpoint(f: typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None], None] = None, *, load_policy: str = 'fresh', temp_dir_root: str = None):
|
912
|
+
"""
|
913
|
+
Enables checkpointing for a step.
|
1082
914
|
|
1083
|
-
- Accessing objects stored in external datastores after task execution.
|
1084
915
|
|
1085
|
-
```python
|
1086
|
-
run = Run("CheckpointsTestsFlow/8992")
|
1087
|
-
with artifact_store_from(run=run, config={
|
1088
|
-
"client_params": {
|
1089
|
-
"aws_access_key_id": os.environ.get("MY_CUSTOM_ACCESS_KEY"),
|
1090
|
-
"aws_secret_access_key": os.environ.get("MY_CUSTOM_SECRET_KEY"),
|
1091
|
-
},
|
1092
|
-
}):
|
1093
|
-
with Checkpoint() as cp:
|
1094
|
-
latest = cp.list(
|
1095
|
-
task=run["start"].task
|
1096
|
-
)[0]
|
1097
|
-
print(latest)
|
1098
|
-
cp.load(
|
1099
|
-
latest,
|
1100
|
-
"test-checkpoints"
|
1101
|
-
)
|
1102
916
|
|
1103
|
-
|
1104
|
-
with artifact_store_from(run=run, config={
|
1105
|
-
"client_params": {
|
1106
|
-
"aws_access_key_id": os.environ.get("MY_CUSTOM_ACCESS_KEY"),
|
1107
|
-
"aws_secret_access_key": os.environ.get("MY_CUSTOM_SECRET_KEY"),
|
1108
|
-
},
|
1109
|
-
}):
|
1110
|
-
load_model(
|
1111
|
-
task.data.model_ref,
|
1112
|
-
"test-models"
|
1113
|
-
)
|
1114
|
-
```
|
1115
|
-
Parameters:
|
917
|
+
Parameters
|
1116
918
|
----------
|
1117
|
-
|
1118
|
-
|
1119
|
-
|
1120
|
-
|
1121
|
-
|
1122
|
-
|
1123
|
-
|
1124
|
-
|
1125
|
-
|
1126
|
-
|
1127
|
-
|
1128
|
-
|
1129
|
-
|
919
|
+
load_policy : str, default: "fresh"
|
920
|
+
The policy for loading the checkpoint. The following policies are supported:
|
921
|
+
- "eager": Loads the the latest available checkpoint within the namespace.
|
922
|
+
With this mode, the latest checkpoint written by any previous task (can be even a different run) of the step
|
923
|
+
will be loaded at the start of the task.
|
924
|
+
- "none": Do not load any checkpoint
|
925
|
+
- "fresh": Loads the lastest checkpoint created within the running Task.
|
926
|
+
This mode helps loading checkpoints across various retry attempts of the same task.
|
927
|
+
With this mode, no checkpoint will be loaded at the start of a task but any checkpoints
|
928
|
+
created within the task will be loaded when the task is retries execution on failure.
|
929
|
+
|
930
|
+
temp_dir_root : str, default: None
|
931
|
+
The root directory under which `current.checkpoint.directory` will be created.
|
1130
932
|
"""
|
1131
933
|
...
|
1132
934
|
|
1133
935
|
@typing.overload
|
1134
|
-
def
|
936
|
+
def catch(*, var: typing.Optional[str] = None, print_exception: bool = True) -> typing.Callable[[typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]], typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]]:
|
1135
937
|
"""
|
1136
|
-
Specifies the
|
938
|
+
Specifies that the step will success under all circumstances.
|
939
|
+
|
940
|
+
The decorator will create an optional artifact, specified by `var`, which
|
941
|
+
contains the exception raised. You can use it to detect the presence
|
942
|
+
of errors, indicating that all happy-path artifacts produced by the step
|
943
|
+
are missing.
|
1137
944
|
|
1138
|
-
Use `@pypi_base` to set common packages required by all
|
1139
|
-
steps and use `@pypi` to specify step-specific overrides.
|
1140
945
|
|
1141
946
|
Parameters
|
1142
947
|
----------
|
1143
|
-
|
1144
|
-
|
1145
|
-
|
1146
|
-
|
1147
|
-
|
1148
|
-
|
948
|
+
var : str, optional, default None
|
949
|
+
Name of the artifact in which to store the caught exception.
|
950
|
+
If not specified, the exception is not stored.
|
951
|
+
print_exception : bool, default True
|
952
|
+
Determines whether or not the exception is printed to
|
953
|
+
stdout when caught.
|
1149
954
|
"""
|
1150
955
|
...
|
1151
956
|
|
1152
957
|
@typing.overload
|
1153
|
-
def
|
958
|
+
def catch(f: typing.Callable[[FlowSpecDerived, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, StepFlag], None]:
|
1154
959
|
...
|
1155
960
|
|
1156
|
-
|
961
|
+
@typing.overload
|
962
|
+
def catch(f: typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]:
|
963
|
+
...
|
964
|
+
|
965
|
+
def catch(f: typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None], None] = None, *, var: typing.Optional[str] = None, print_exception: bool = True):
|
1157
966
|
"""
|
1158
|
-
Specifies the
|
967
|
+
Specifies that the step will success under all circumstances.
|
968
|
+
|
969
|
+
The decorator will create an optional artifact, specified by `var`, which
|
970
|
+
contains the exception raised. You can use it to detect the presence
|
971
|
+
of errors, indicating that all happy-path artifacts produced by the step
|
972
|
+
are missing.
|
1159
973
|
|
1160
|
-
Use `@pypi_base` to set common packages required by all
|
1161
|
-
steps and use `@pypi` to specify step-specific overrides.
|
1162
974
|
|
1163
975
|
Parameters
|
1164
976
|
----------
|
1165
|
-
|
1166
|
-
|
1167
|
-
|
1168
|
-
|
1169
|
-
|
1170
|
-
|
977
|
+
var : str, optional, default None
|
978
|
+
Name of the artifact in which to store the caught exception.
|
979
|
+
If not specified, the exception is not stored.
|
980
|
+
print_exception : bool, default True
|
981
|
+
Determines whether or not the exception is printed to
|
982
|
+
stdout when caught.
|
1171
983
|
"""
|
1172
984
|
...
|
1173
985
|
|
1174
|
-
|
986
|
+
@typing.overload
|
987
|
+
def environment(*, vars: typing.Dict[str, str] = {}) -> typing.Callable[[typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]], typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]]]:
|
1175
988
|
"""
|
1176
|
-
Specifies
|
1177
|
-
|
1178
|
-
A project-specific namespace is created for all flows that
|
1179
|
-
use the same `@project(name)`.
|
989
|
+
Specifies environment variables to be set prior to the execution of a step.
|
1180
990
|
|
1181
991
|
|
1182
992
|
Parameters
|
1183
993
|
----------
|
1184
|
-
|
1185
|
-
|
1186
|
-
projects that use the same production scheduler. The name may
|
1187
|
-
contain only lowercase alphanumeric characters and underscores.
|
1188
|
-
|
1189
|
-
branch : Optional[str], default None
|
1190
|
-
The branch to use. If not specified, the branch is set to
|
1191
|
-
`user.<username>` unless `production` is set to `True`. This can
|
1192
|
-
also be set on the command line using `--branch` as a top-level option.
|
1193
|
-
It is an error to specify `branch` in the decorator and on the command line.
|
1194
|
-
|
1195
|
-
production : bool, default False
|
1196
|
-
Whether or not the branch is the production branch. This can also be set on the
|
1197
|
-
command line using `--production` as a top-level option. It is an error to specify
|
1198
|
-
`production` in the decorator and on the command line.
|
1199
|
-
The project branch name will be:
|
1200
|
-
- if `branch` is specified:
|
1201
|
-
- if `production` is True: `prod.<branch>`
|
1202
|
-
- if `production` is False: `test.<branch>`
|
1203
|
-
- if `branch` is not specified:
|
1204
|
-
- if `production` is True: `prod`
|
1205
|
-
- if `production` is False: `user.<username>`
|
994
|
+
vars : Dict[str, str], default {}
|
995
|
+
Dictionary of environment variables to set.
|
1206
996
|
"""
|
1207
997
|
...
|
1208
998
|
|
1209
|
-
|
999
|
+
@typing.overload
|
1000
|
+
def environment(f: typing.Callable[[FlowSpecDerived, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, StepFlag], None]:
|
1001
|
+
...
|
1002
|
+
|
1003
|
+
@typing.overload
|
1004
|
+
def environment(f: typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]) -> typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None]:
|
1005
|
+
...
|
1006
|
+
|
1007
|
+
def environment(f: typing.Union[typing.Callable[[FlowSpecDerived, StepFlag], None], typing.Callable[[FlowSpecDerived, typing.Any, StepFlag], None], None] = None, *, vars: typing.Dict[str, str] = {}):
|
1210
1008
|
"""
|
1211
|
-
|
1212
|
-
This decorator only works when a flow is scheduled on Airflow and is compiled using `airflow create`. More than one `@airflow_external_task_sensor` can be added as a flow decorators. Adding more than one decorator will ensure that `start` step starts only after all sensors finish.
|
1009
|
+
Specifies environment variables to be set prior to the execution of a step.
|
1213
1010
|
|
1214
1011
|
|
1215
1012
|
Parameters
|
1216
1013
|
----------
|
1217
|
-
|
1218
|
-
|
1219
|
-
poke_interval : int
|
1220
|
-
Time in seconds that the job should wait in between each try. (Default: 60)
|
1221
|
-
mode : str
|
1222
|
-
How the sensor operates. Options are: { poke | reschedule }. (Default: "poke")
|
1223
|
-
exponential_backoff : bool
|
1224
|
-
allow progressive longer waits between pokes by using exponential backoff algorithm. (Default: True)
|
1225
|
-
pool : str
|
1226
|
-
the slot pool this task should run in,
|
1227
|
-
slot pools are a way to limit concurrency for certain tasks. (Default:None)
|
1228
|
-
soft_fail : bool
|
1229
|
-
Set to true to mark the task as SKIPPED on failure. (Default: False)
|
1230
|
-
name : str
|
1231
|
-
Name of the sensor on Airflow
|
1232
|
-
description : str
|
1233
|
-
Description of sensor in the Airflow UI
|
1234
|
-
external_dag_id : str
|
1235
|
-
The dag_id that contains the task you want to wait for.
|
1236
|
-
external_task_ids : List[str]
|
1237
|
-
The list of task_ids that you want to wait for.
|
1238
|
-
If None (default value) the sensor waits for the DAG. (Default: None)
|
1239
|
-
allowed_states : List[str]
|
1240
|
-
Iterable of allowed states, (Default: ['success'])
|
1241
|
-
failed_states : List[str]
|
1242
|
-
Iterable of failed or dis-allowed states. (Default: None)
|
1243
|
-
execution_delta : datetime.timedelta
|
1244
|
-
time difference with the previous execution to look at,
|
1245
|
-
the default is the same logical date as the current task or DAG. (Default: None)
|
1246
|
-
check_existence: bool
|
1247
|
-
Set to True to check if the external task exists or check if
|
1248
|
-
the DAG to wait for exists. (Default: True)
|
1014
|
+
vars : Dict[str, str], default {}
|
1015
|
+
Dictionary of environment variables to set.
|
1249
1016
|
"""
|
1250
1017
|
...
|
1251
1018
|
|
@@ -1342,54 +1109,287 @@ def trigger(f: typing.Optional[typing.Type[FlowSpecDerived]] = None, *, event: t
|
|
1342
1109
|
"""
|
1343
1110
|
...
|
1344
1111
|
|
1112
|
+
def airflow_s3_key_sensor(*, timeout: int, poke_interval: int, mode: str, exponential_backoff: bool, pool: str, soft_fail: bool, name: str, description: str, bucket_key: typing.Union[str, typing.List[str]], bucket_name: str, wildcard_match: bool, aws_conn_id: str, verify: bool) -> typing.Callable[[typing.Type[FlowSpecDerived]], typing.Type[FlowSpecDerived]]:
|
1113
|
+
"""
|
1114
|
+
The `@airflow_s3_key_sensor` decorator attaches a Airflow [S3KeySensor](https://airflow.apache.org/docs/apache-airflow-providers-amazon/stable/_api/airflow/providers/amazon/aws/sensors/s3/index.html#airflow.providers.amazon.aws.sensors.s3.S3KeySensor)
|
1115
|
+
before the start step of the flow. This decorator only works when a flow is scheduled on Airflow
|
1116
|
+
and is compiled using `airflow create`. More than one `@airflow_s3_key_sensor` can be
|
1117
|
+
added as a flow decorators. Adding more than one decorator will ensure that `start` step
|
1118
|
+
starts only after all sensors finish.
|
1119
|
+
|
1120
|
+
|
1121
|
+
Parameters
|
1122
|
+
----------
|
1123
|
+
timeout : int
|
1124
|
+
Time, in seconds before the task times out and fails. (Default: 3600)
|
1125
|
+
poke_interval : int
|
1126
|
+
Time in seconds that the job should wait in between each try. (Default: 60)
|
1127
|
+
mode : str
|
1128
|
+
How the sensor operates. Options are: { poke | reschedule }. (Default: "poke")
|
1129
|
+
exponential_backoff : bool
|
1130
|
+
allow progressive longer waits between pokes by using exponential backoff algorithm. (Default: True)
|
1131
|
+
pool : str
|
1132
|
+
the slot pool this task should run in,
|
1133
|
+
slot pools are a way to limit concurrency for certain tasks. (Default:None)
|
1134
|
+
soft_fail : bool
|
1135
|
+
Set to true to mark the task as SKIPPED on failure. (Default: False)
|
1136
|
+
name : str
|
1137
|
+
Name of the sensor on Airflow
|
1138
|
+
description : str
|
1139
|
+
Description of sensor in the Airflow UI
|
1140
|
+
bucket_key : Union[str, List[str]]
|
1141
|
+
The key(s) being waited on. Supports full s3:// style url or relative path from root level.
|
1142
|
+
When it's specified as a full s3:// url, please leave `bucket_name` as None
|
1143
|
+
bucket_name : str
|
1144
|
+
Name of the S3 bucket. Only needed when bucket_key is not provided as a full s3:// url.
|
1145
|
+
When specified, all the keys passed to bucket_key refers to this bucket. (Default:None)
|
1146
|
+
wildcard_match : bool
|
1147
|
+
whether the bucket_key should be interpreted as a Unix wildcard pattern. (Default: False)
|
1148
|
+
aws_conn_id : str
|
1149
|
+
a reference to the s3 connection on Airflow. (Default: None)
|
1150
|
+
verify : bool
|
1151
|
+
Whether or not to verify SSL certificates for S3 connection. (Default: None)
|
1152
|
+
"""
|
1153
|
+
...
|
1154
|
+
|
1155
|
+
@typing.overload
|
1156
|
+
def schedule(*, hourly: bool = False, daily: bool = True, weekly: bool = False, cron: typing.Optional[str] = None, timezone: typing.Optional[str] = None) -> typing.Callable[[typing.Type[FlowSpecDerived]], typing.Type[FlowSpecDerived]]:
|
1157
|
+
"""
|
1158
|
+
Specifies the times when the flow should be run when running on a
|
1159
|
+
production scheduler.
|
1160
|
+
|
1161
|
+
|
1162
|
+
Parameters
|
1163
|
+
----------
|
1164
|
+
hourly : bool, default False
|
1165
|
+
Run the workflow hourly.
|
1166
|
+
daily : bool, default True
|
1167
|
+
Run the workflow daily.
|
1168
|
+
weekly : bool, default False
|
1169
|
+
Run the workflow weekly.
|
1170
|
+
cron : str, optional, default None
|
1171
|
+
Run the workflow at [a custom Cron schedule](https://docs.aws.amazon.com/eventbridge/latest/userguide/scheduled-events.html#cron-expressions)
|
1172
|
+
specified by this expression.
|
1173
|
+
timezone : str, optional, default None
|
1174
|
+
Timezone on which the schedule runs (default: None). Currently supported only for Argo workflows,
|
1175
|
+
which accepts timezones in [IANA format](https://nodatime.org/TimeZones).
|
1176
|
+
"""
|
1177
|
+
...
|
1178
|
+
|
1179
|
+
@typing.overload
|
1180
|
+
def schedule(f: typing.Type[FlowSpecDerived]) -> typing.Type[FlowSpecDerived]:
|
1181
|
+
...
|
1182
|
+
|
1183
|
+
def schedule(f: typing.Optional[typing.Type[FlowSpecDerived]] = None, *, hourly: bool = False, daily: bool = True, weekly: bool = False, cron: typing.Optional[str] = None, timezone: typing.Optional[str] = None):
|
1184
|
+
"""
|
1185
|
+
Specifies the times when the flow should be run when running on a
|
1186
|
+
production scheduler.
|
1187
|
+
|
1188
|
+
|
1189
|
+
Parameters
|
1190
|
+
----------
|
1191
|
+
hourly : bool, default False
|
1192
|
+
Run the workflow hourly.
|
1193
|
+
daily : bool, default True
|
1194
|
+
Run the workflow daily.
|
1195
|
+
weekly : bool, default False
|
1196
|
+
Run the workflow weekly.
|
1197
|
+
cron : str, optional, default None
|
1198
|
+
Run the workflow at [a custom Cron schedule](https://docs.aws.amazon.com/eventbridge/latest/userguide/scheduled-events.html#cron-expressions)
|
1199
|
+
specified by this expression.
|
1200
|
+
timezone : str, optional, default None
|
1201
|
+
Timezone on which the schedule runs (default: None). Currently supported only for Argo workflows,
|
1202
|
+
which accepts timezones in [IANA format](https://nodatime.org/TimeZones).
|
1203
|
+
"""
|
1204
|
+
...
|
1205
|
+
|
1206
|
+
def project(*, name: str, branch: typing.Optional[str] = None, production: bool = False) -> typing.Callable[[typing.Type[FlowSpecDerived]], typing.Type[FlowSpecDerived]]:
|
1207
|
+
"""
|
1208
|
+
Specifies what flows belong to the same project.
|
1209
|
+
|
1210
|
+
A project-specific namespace is created for all flows that
|
1211
|
+
use the same `@project(name)`.
|
1212
|
+
|
1213
|
+
|
1214
|
+
Parameters
|
1215
|
+
----------
|
1216
|
+
name : str
|
1217
|
+
Project name. Make sure that the name is unique amongst all
|
1218
|
+
projects that use the same production scheduler. The name may
|
1219
|
+
contain only lowercase alphanumeric characters and underscores.
|
1220
|
+
|
1221
|
+
branch : Optional[str], default None
|
1222
|
+
The branch to use. If not specified, the branch is set to
|
1223
|
+
`user.<username>` unless `production` is set to `True`. This can
|
1224
|
+
also be set on the command line using `--branch` as a top-level option.
|
1225
|
+
It is an error to specify `branch` in the decorator and on the command line.
|
1226
|
+
|
1227
|
+
production : bool, default False
|
1228
|
+
Whether or not the branch is the production branch. This can also be set on the
|
1229
|
+
command line using `--production` as a top-level option. It is an error to specify
|
1230
|
+
`production` in the decorator and on the command line.
|
1231
|
+
The project branch name will be:
|
1232
|
+
- if `branch` is specified:
|
1233
|
+
- if `production` is True: `prod.<branch>`
|
1234
|
+
- if `production` is False: `test.<branch>`
|
1235
|
+
- if `branch` is not specified:
|
1236
|
+
- if `production` is True: `prod`
|
1237
|
+
- if `production` is False: `user.<username>`
|
1238
|
+
"""
|
1239
|
+
...
|
1240
|
+
|
1345
1241
|
@typing.overload
|
1346
|
-
def
|
1242
|
+
def pypi_base(*, packages: typing.Dict[str, str] = {}, python: typing.Optional[str] = None) -> typing.Callable[[typing.Type[FlowSpecDerived]], typing.Type[FlowSpecDerived]]:
|
1347
1243
|
"""
|
1348
|
-
Specifies the
|
1349
|
-
|
1350
|
-
Use `@conda_base` to set common libraries required by all
|
1351
|
-
steps and use `@conda` to specify step-specific additions.
|
1244
|
+
Specifies the PyPI packages for all steps of the flow.
|
1352
1245
|
|
1246
|
+
Use `@pypi_base` to set common packages required by all
|
1247
|
+
steps and use `@pypi` to specify step-specific overrides.
|
1353
1248
|
|
1354
1249
|
Parameters
|
1355
1250
|
----------
|
1356
|
-
packages : Dict[str, str], default {}
|
1251
|
+
packages : Dict[str, str], default: {}
|
1357
1252
|
Packages to use for this flow. The key is the name of the package
|
1358
1253
|
and the value is the version to use.
|
1359
|
-
|
1360
|
-
Supported for backward compatibility. When used with packages, packages will take precedence.
|
1361
|
-
python : str, optional, default None
|
1254
|
+
python : str, optional, default: None
|
1362
1255
|
Version of Python to use, e.g. '3.7.4'. A default value of None implies
|
1363
1256
|
that the version used will correspond to the version of the Python interpreter used to start the run.
|
1364
|
-
disabled : bool, default False
|
1365
|
-
If set to True, disables Conda.
|
1366
1257
|
"""
|
1367
1258
|
...
|
1368
1259
|
|
1369
1260
|
@typing.overload
|
1370
|
-
def
|
1261
|
+
def pypi_base(f: typing.Type[FlowSpecDerived]) -> typing.Type[FlowSpecDerived]:
|
1371
1262
|
...
|
1372
1263
|
|
1373
|
-
def
|
1264
|
+
def pypi_base(f: typing.Optional[typing.Type[FlowSpecDerived]] = None, *, packages: typing.Dict[str, str] = {}, python: typing.Optional[str] = None):
|
1374
1265
|
"""
|
1375
|
-
Specifies the
|
1376
|
-
|
1377
|
-
Use `@conda_base` to set common libraries required by all
|
1378
|
-
steps and use `@conda` to specify step-specific additions.
|
1266
|
+
Specifies the PyPI packages for all steps of the flow.
|
1379
1267
|
|
1268
|
+
Use `@pypi_base` to set common packages required by all
|
1269
|
+
steps and use `@pypi` to specify step-specific overrides.
|
1380
1270
|
|
1381
1271
|
Parameters
|
1382
1272
|
----------
|
1383
|
-
packages : Dict[str, str], default {}
|
1273
|
+
packages : Dict[str, str], default: {}
|
1384
1274
|
Packages to use for this flow. The key is the name of the package
|
1385
1275
|
and the value is the version to use.
|
1386
|
-
|
1387
|
-
Supported for backward compatibility. When used with packages, packages will take precedence.
|
1388
|
-
python : str, optional, default None
|
1276
|
+
python : str, optional, default: None
|
1389
1277
|
Version of Python to use, e.g. '3.7.4'. A default value of None implies
|
1390
1278
|
that the version used will correspond to the version of the Python interpreter used to start the run.
|
1391
|
-
|
1392
|
-
|
1279
|
+
"""
|
1280
|
+
...
|
1281
|
+
|
1282
|
+
def with_artifact_store(f: typing.Optional[typing.Type[FlowSpecDerived]] = None):
|
1283
|
+
"""
|
1284
|
+
Allows setting external datastores to save data for the
|
1285
|
+
`@checkpoint`/`@model`/`@huggingface_hub` decorators.
|
1286
|
+
|
1287
|
+
This decorator is useful when users wish to save data to a different datastore
|
1288
|
+
than what is configured in Metaflow. This can be for variety of reasons:
|
1289
|
+
|
1290
|
+
1. Data security: The objects needs to be stored in a bucket (object storage) that is not accessible by other flows.
|
1291
|
+
2. Data Locality: The location where the task is executing is not located in the same region as the datastore.
|
1292
|
+
- Example: Metaflow datastore lives in US East, but the task is executing in Finland datacenters.
|
1293
|
+
3. Data Lifecycle Policies: The objects need to be archived / managed separately from the Metaflow managed objects.
|
1294
|
+
- Example: Flow is training very large models that need to be stored separately and will be deleted more aggressively than the Metaflow managed objects.
|
1295
|
+
|
1296
|
+
Usage:
|
1297
|
+
----------
|
1298
|
+
|
1299
|
+
- Using a custom IAM role to access the datastore.
|
1300
|
+
|
1301
|
+
```python
|
1302
|
+
@with_artifact_store(
|
1303
|
+
type="s3",
|
1304
|
+
config=lambda: {
|
1305
|
+
"root": "s3://my-bucket-foo/path/to/root",
|
1306
|
+
"role_arn": ROLE,
|
1307
|
+
},
|
1308
|
+
)
|
1309
|
+
class MyFlow(FlowSpec):
|
1310
|
+
|
1311
|
+
@checkpoint
|
1312
|
+
@step
|
1313
|
+
def start(self):
|
1314
|
+
with open("my_file.txt", "w") as f:
|
1315
|
+
f.write("Hello, World!")
|
1316
|
+
self.external_bucket_checkpoint = current.checkpoint.save("my_file.txt")
|
1317
|
+
self.next(self.end)
|
1318
|
+
|
1319
|
+
```
|
1320
|
+
|
1321
|
+
- Using credentials to access the s3-compatible datastore.
|
1322
|
+
|
1323
|
+
```python
|
1324
|
+
@with_artifact_store(
|
1325
|
+
type="s3",
|
1326
|
+
config=lambda: {
|
1327
|
+
"root": "s3://my-bucket-foo/path/to/root",
|
1328
|
+
"client_params": {
|
1329
|
+
"aws_access_key_id": os.environ.get("MY_CUSTOM_ACCESS_KEY"),
|
1330
|
+
"aws_secret_access_key": os.environ.get("MY_CUSTOM_SECRET_KEY"),
|
1331
|
+
},
|
1332
|
+
},
|
1333
|
+
)
|
1334
|
+
class MyFlow(FlowSpec):
|
1335
|
+
|
1336
|
+
@checkpoint
|
1337
|
+
@step
|
1338
|
+
def start(self):
|
1339
|
+
with open("my_file.txt", "w") as f:
|
1340
|
+
f.write("Hello, World!")
|
1341
|
+
self.external_bucket_checkpoint = current.checkpoint.save("my_file.txt")
|
1342
|
+
self.next(self.end)
|
1343
|
+
|
1344
|
+
```
|
1345
|
+
|
1346
|
+
- Accessing objects stored in external datastores after task execution.
|
1347
|
+
|
1348
|
+
```python
|
1349
|
+
run = Run("CheckpointsTestsFlow/8992")
|
1350
|
+
with artifact_store_from(run=run, config={
|
1351
|
+
"client_params": {
|
1352
|
+
"aws_access_key_id": os.environ.get("MY_CUSTOM_ACCESS_KEY"),
|
1353
|
+
"aws_secret_access_key": os.environ.get("MY_CUSTOM_SECRET_KEY"),
|
1354
|
+
},
|
1355
|
+
}):
|
1356
|
+
with Checkpoint() as cp:
|
1357
|
+
latest = cp.list(
|
1358
|
+
task=run["start"].task
|
1359
|
+
)[0]
|
1360
|
+
print(latest)
|
1361
|
+
cp.load(
|
1362
|
+
latest,
|
1363
|
+
"test-checkpoints"
|
1364
|
+
)
|
1365
|
+
|
1366
|
+
task = Task("TorchTuneFlow/8484/train/53673")
|
1367
|
+
with artifact_store_from(run=run, config={
|
1368
|
+
"client_params": {
|
1369
|
+
"aws_access_key_id": os.environ.get("MY_CUSTOM_ACCESS_KEY"),
|
1370
|
+
"aws_secret_access_key": os.environ.get("MY_CUSTOM_SECRET_KEY"),
|
1371
|
+
},
|
1372
|
+
}):
|
1373
|
+
load_model(
|
1374
|
+
task.data.model_ref,
|
1375
|
+
"test-models"
|
1376
|
+
)
|
1377
|
+
```
|
1378
|
+
Parameters:
|
1379
|
+
----------
|
1380
|
+
|
1381
|
+
type: str
|
1382
|
+
The type of the datastore. Can be one of 's3', 'gcs', 'azure' or any other supported metaflow Datastore.
|
1383
|
+
|
1384
|
+
config: dict or Callable
|
1385
|
+
Dictionary of configuration options for the datastore. The following keys are required:
|
1386
|
+
- root: The root path in the datastore where the data will be saved. (needs to be in the format expected by the datastore)
|
1387
|
+
- example: 's3://bucket-name/path/to/root'
|
1388
|
+
- example: 'gs://bucket-name/path/to/root'
|
1389
|
+
- example: 'https://myblockacc.blob.core.windows.net/metaflow/'
|
1390
|
+
- role_arn (optional): AWS IAM role to access s3 bucket (only when `type` is 's3')
|
1391
|
+
- session_vars (optional): AWS session variables to access s3 bucket (only when `type` is 's3')
|
1392
|
+
- client_params (optional): AWS client parameters to access s3 bucket (only when `type` is 's3')
|
1393
1393
|
"""
|
1394
1394
|
...
|
1395
1395
|
|
@@ -1495,63 +1495,60 @@ def trigger_on_finish(f: typing.Optional[typing.Type[FlowSpecDerived]] = None, *
|
|
1495
1495
|
...
|
1496
1496
|
|
1497
1497
|
@typing.overload
|
1498
|
-
def
|
1498
|
+
def conda_base(*, packages: typing.Dict[str, str] = {}, libraries: typing.Dict[str, str] = {}, python: typing.Optional[str] = None, disabled: bool = False) -> typing.Callable[[typing.Type[FlowSpecDerived]], typing.Type[FlowSpecDerived]]:
|
1499
1499
|
"""
|
1500
|
-
Specifies the
|
1501
|
-
|
1500
|
+
Specifies the Conda environment for all steps of the flow.
|
1501
|
+
|
1502
|
+
Use `@conda_base` to set common libraries required by all
|
1503
|
+
steps and use `@conda` to specify step-specific additions.
|
1502
1504
|
|
1503
1505
|
|
1504
1506
|
Parameters
|
1505
1507
|
----------
|
1506
|
-
|
1507
|
-
|
1508
|
-
|
1509
|
-
|
1510
|
-
|
1511
|
-
|
1512
|
-
|
1513
|
-
|
1514
|
-
|
1515
|
-
|
1516
|
-
Timezone on which the schedule runs (default: None). Currently supported only for Argo workflows,
|
1517
|
-
which accepts timezones in [IANA format](https://nodatime.org/TimeZones).
|
1508
|
+
packages : Dict[str, str], default {}
|
1509
|
+
Packages to use for this flow. The key is the name of the package
|
1510
|
+
and the value is the version to use.
|
1511
|
+
libraries : Dict[str, str], default {}
|
1512
|
+
Supported for backward compatibility. When used with packages, packages will take precedence.
|
1513
|
+
python : str, optional, default None
|
1514
|
+
Version of Python to use, e.g. '3.7.4'. A default value of None implies
|
1515
|
+
that the version used will correspond to the version of the Python interpreter used to start the run.
|
1516
|
+
disabled : bool, default False
|
1517
|
+
If set to True, disables Conda.
|
1518
1518
|
"""
|
1519
1519
|
...
|
1520
1520
|
|
1521
1521
|
@typing.overload
|
1522
|
-
def
|
1522
|
+
def conda_base(f: typing.Type[FlowSpecDerived]) -> typing.Type[FlowSpecDerived]:
|
1523
1523
|
...
|
1524
1524
|
|
1525
|
-
def
|
1525
|
+
def conda_base(f: typing.Optional[typing.Type[FlowSpecDerived]] = None, *, packages: typing.Dict[str, str] = {}, libraries: typing.Dict[str, str] = {}, python: typing.Optional[str] = None, disabled: bool = False):
|
1526
1526
|
"""
|
1527
|
-
Specifies the
|
1528
|
-
|
1527
|
+
Specifies the Conda environment for all steps of the flow.
|
1528
|
+
|
1529
|
+
Use `@conda_base` to set common libraries required by all
|
1530
|
+
steps and use `@conda` to specify step-specific additions.
|
1529
1531
|
|
1530
1532
|
|
1531
1533
|
Parameters
|
1532
1534
|
----------
|
1533
|
-
|
1534
|
-
|
1535
|
-
|
1536
|
-
|
1537
|
-
|
1538
|
-
|
1539
|
-
|
1540
|
-
|
1541
|
-
|
1542
|
-
|
1543
|
-
Timezone on which the schedule runs (default: None). Currently supported only for Argo workflows,
|
1544
|
-
which accepts timezones in [IANA format](https://nodatime.org/TimeZones).
|
1535
|
+
packages : Dict[str, str], default {}
|
1536
|
+
Packages to use for this flow. The key is the name of the package
|
1537
|
+
and the value is the version to use.
|
1538
|
+
libraries : Dict[str, str], default {}
|
1539
|
+
Supported for backward compatibility. When used with packages, packages will take precedence.
|
1540
|
+
python : str, optional, default None
|
1541
|
+
Version of Python to use, e.g. '3.7.4'. A default value of None implies
|
1542
|
+
that the version used will correspond to the version of the Python interpreter used to start the run.
|
1543
|
+
disabled : bool, default False
|
1544
|
+
If set to True, disables Conda.
|
1545
1545
|
"""
|
1546
1546
|
...
|
1547
1547
|
|
1548
|
-
def
|
1548
|
+
def airflow_external_task_sensor(*, timeout: int, poke_interval: int, mode: str, exponential_backoff: bool, pool: str, soft_fail: bool, name: str, description: str, external_dag_id: str, external_task_ids: typing.List[str], allowed_states: typing.List[str], failed_states: typing.List[str], execution_delta: "datetime.timedelta", check_existence: bool) -> typing.Callable[[typing.Type[FlowSpecDerived]], typing.Type[FlowSpecDerived]]:
|
1549
1549
|
"""
|
1550
|
-
The `@
|
1551
|
-
|
1552
|
-
and is compiled using `airflow create`. More than one `@airflow_s3_key_sensor` can be
|
1553
|
-
added as a flow decorators. Adding more than one decorator will ensure that `start` step
|
1554
|
-
starts only after all sensors finish.
|
1550
|
+
The `@airflow_external_task_sensor` decorator attaches a Airflow [ExternalTaskSensor](https://airflow.apache.org/docs/apache-airflow/stable/_api/airflow/sensors/external_task/index.html#airflow.sensors.external_task.ExternalTaskSensor) before the start step of the flow.
|
1551
|
+
This decorator only works when a flow is scheduled on Airflow and is compiled using `airflow create`. More than one `@airflow_external_task_sensor` can be added as a flow decorators. Adding more than one decorator will ensure that `start` step starts only after all sensors finish.
|
1555
1552
|
|
1556
1553
|
|
1557
1554
|
Parameters
|
@@ -1573,18 +1570,21 @@ def airflow_s3_key_sensor(*, timeout: int, poke_interval: int, mode: str, expone
|
|
1573
1570
|
Name of the sensor on Airflow
|
1574
1571
|
description : str
|
1575
1572
|
Description of sensor in the Airflow UI
|
1576
|
-
|
1577
|
-
The
|
1578
|
-
|
1579
|
-
|
1580
|
-
|
1581
|
-
|
1582
|
-
|
1583
|
-
|
1584
|
-
|
1585
|
-
|
1586
|
-
|
1587
|
-
|
1573
|
+
external_dag_id : str
|
1574
|
+
The dag_id that contains the task you want to wait for.
|
1575
|
+
external_task_ids : List[str]
|
1576
|
+
The list of task_ids that you want to wait for.
|
1577
|
+
If None (default value) the sensor waits for the DAG. (Default: None)
|
1578
|
+
allowed_states : List[str]
|
1579
|
+
Iterable of allowed states, (Default: ['success'])
|
1580
|
+
failed_states : List[str]
|
1581
|
+
Iterable of failed or dis-allowed states. (Default: None)
|
1582
|
+
execution_delta : datetime.timedelta
|
1583
|
+
time difference with the previous execution to look at,
|
1584
|
+
the default is the same logical date as the current task or DAG. (Default: None)
|
1585
|
+
check_existence: bool
|
1586
|
+
Set to True to check if the external task exists or check if
|
1587
|
+
the DAG to wait for exists. (Default: True)
|
1588
1588
|
"""
|
1589
1589
|
...
|
1590
1590
|
|