Module sft (1.122.0)

Classes for supervised tuning.

Classes

SupervisedTuningJob

SupervisedTuningJob(tuning_job_name: str)

Initializes class with project, location, and api_client.

Parameters
Name	Description
`project`	`str` Project of the resource noun.
`location`	`str` The location of the resource noun.
`credentials`	`google.auth.credentials.Credentials` Optional custom credentials to use when accessing interacting with resource noun.
`resource_name`	`str` A fully-qualified resource name or ID.

Modules Functions

preview_train

preview_train(
 *,
 source_model: typing.Union[
 str,
 vertexai.generative_models.GenerativeModel,
 vertexai.preview.tuning._tuning.SourceModel,
 ],
 train_dataset: typing.Union[
 str, google.cloud.aiplatform.preview.datasets.MultimodalDataset
 ],
 validation_dataset: typing.Optional[
 typing.Union[str, google.cloud.aiplatform.preview.datasets.MultimodalDataset]
 ] = None,
 tuned_model_display_name: typing.Optional[str] = None,
 tuning_mode: typing.Optional[typing.Literal["FULL", "PEFT_ADAPTER"]] = None,
 epochs: typing.Optional[int] = None,
 learning_rate: typing.Optional[float] = None,
 learning_rate_multiplier: typing.Optional[float] = None,
 adapter_size: typing.Optional[typing.Literal[1, 4, 8, 16, 32]] = None,
 labels: typing.Optional[typing.Dict[str, str]] = None,
 output_uri: typing.Optional[str] = None
) -> vertexai.preview.tuning._supervised_tuning.SupervisedTuningJob

Tunes a model using supervised training.

Parameter
Name	Description
`source_model`	`str` Model name for tuning, e.g., "gemini-1.0-pro-002".

rebase_tuned_model

rebase_tuned_model(
 tuned_model_ref: str,
 *,
 artifact_destination: typing.Optional[str] = None,
 deploy_to_same_endpoint: typing.Optional[bool] = False
)

Re-runs fine tuning on top of a new foundational model.

Takes a legacy Tuned GenAI model Reference and creates a TuningJob based on a new model.

train

train(
 *,
 source_model: typing.Union[str, vertexai.generative_models.GenerativeModel],
 train_dataset: typing.Union[
 str, google.cloud.aiplatform.preview.datasets.MultimodalDataset
 ],
 validation_dataset: typing.Optional[
 typing.Union[str, google.cloud.aiplatform.preview.datasets.MultimodalDataset]
 ] = None,
 tuned_model_display_name: typing.Optional[str] = None,
 epochs: typing.Optional[int] = None,
 learning_rate_multiplier: typing.Optional[float] = None,
 adapter_size: typing.Optional[typing.Literal[1, 4, 8, 16, 32]] = None,
 labels: typing.Optional[typing.Dict[str, str]] = None
) -> vertexai.tuning._supervised_tuning.SupervisedTuningJob