Class TextGenerationModel (1.60.0)

TextGenerationModel(model_id: str, endpoint_name: typing.Optional[str] = None)

Creates a LanguageModel.

This constructor should not be called directly. Use LanguageModel.from_pretrained(model_name=...) instead.

Parameters
Name	Description
`model_id`	`str` Identifier of a Vertex LLM. Example: "text-bison@001"
`endpoint_name`	`typing.Optional[str]` Vertex Endpoint resource name for the model

Methods

batch_predict

batch_predict(
 *,
 dataset: typing.Union[str, typing.List[str]],
 destination_uri_prefix: str,
 model_parameters: typing.Optional[typing.Dict] = None
) -> google.cloud.aiplatform.jobs.BatchPredictionJob

Starts a batch prediction job with the model.

Exceptions
Type	Description
`ValueError`	When source or destination URI is not supported.

from_pretrained

from_pretrained(model_name: str) -> vertexai._model_garden._model_garden_models.T

Loads a _ModelGardenModel.

Parameter
Name	Description
`model_name`	`str` Name of the model.

Exceptions
Type	Description
`ValueError`	If model_name is unknown.
`ValueError`	If model does not support this class.

get_tuned_model

get_tuned_model(
 tuned_model_name: str,
) -> vertexai.language_models._language_models._LanguageModel

Loads the specified tuned language model.

list_tuned_model_names

list_tuned_model_names() -> typing.Sequence[str]

Lists the names of tuned models.

predict

predict(
 prompt: str,
 *,
 max_output_tokens: typing.Optional[int] = 128,
 temperature: typing.Optional[float] = None,
 top_k: typing.Optional[int] = None,
 top_p: typing.Optional[float] = None,
 stop_sequences: typing.Optional[typing.List[str]] = None,
 candidate_count: typing.Optional[int] = None,
 grounding_source: typing.Optional[
 typing.Union[
 vertexai.language_models._language_models.WebSearch,
 vertexai.language_models._language_models.VertexAISearch,
 vertexai.language_models._language_models.InlineContext,
 ]
 ] = None,
 logprobs: typing.Optional[int] = None,
 presence_penalty: typing.Optional[float] = None,
 frequency_penalty: typing.Optional[float] = None,
 logit_bias: typing.Optional[typing.Dict[str, float]] = None,
 seed: typing.Optional[int] = None
) -> vertexai.language_models.MultiCandidateTextGenerationResponse

Gets model response for a single prompt.

Parameter
Name	Description
`prompt`	`str` Question to ask the model.

predict_async

predict_async(
 prompt: str,
 *,
 max_output_tokens: typing.Optional[int] = 128,
 temperature: typing.Optional[float] = None,
 top_k: typing.Optional[int] = None,
 top_p: typing.Optional[float] = None,
 stop_sequences: typing.Optional[typing.List[str]] = None,
 candidate_count: typing.Optional[int] = None,
 grounding_source: typing.Optional[
 typing.Union[
 vertexai.language_models._language_models.WebSearch,
 vertexai.language_models._language_models.VertexAISearch,
 vertexai.language_models._language_models.InlineContext,
 ]
 ] = None,
 logprobs: typing.Optional[int] = None,
 presence_penalty: typing.Optional[float] = None,
 frequency_penalty: typing.Optional[float] = None,
 logit_bias: typing.Optional[typing.Dict[str, float]] = None,
 seed: typing.Optional[int] = None
) -> vertexai.language_models.MultiCandidateTextGenerationResponse

Asynchronously gets model response for a single prompt.

Parameter
Name	Description
`prompt`	`str` Question to ask the model.

predict_streaming

predict_streaming(
 prompt: str,
 *,
 max_output_tokens: int = 128,
 temperature: typing.Optional[float] = None,
 top_k: typing.Optional[int] = None,
 top_p: typing.Optional[float] = None,
 stop_sequences: typing.Optional[typing.List[str]] = None,
 logprobs: typing.Optional[int] = None,
 presence_penalty: typing.Optional[float] = None,
 frequency_penalty: typing.Optional[float] = None,
 logit_bias: typing.Optional[typing.Dict[str, float]] = None,
 seed: typing.Optional[int] = None
) -> typing.Iterator[vertexai.language_models.TextGenerationResponse]

Gets a streaming model response for a single prompt.

The result is a stream (generator) of partial responses.

Parameter
Name	Description
`prompt`	`str` Question to ask the model.

predict_streaming_async

predict_streaming_async(
 prompt: str,
 *,
 max_output_tokens: int = 128,
 temperature: typing.Optional[float] = None,
 top_k: typing.Optional[int] = None,
 top_p: typing.Optional[float] = None,
 stop_sequences: typing.Optional[typing.List[str]] = None,
 logprobs: typing.Optional[int] = None,
 presence_penalty: typing.Optional[float] = None,
 frequency_penalty: typing.Optional[float] = None,
 logit_bias: typing.Optional[typing.Dict[str, float]] = None,
 seed: typing.Optional[int] = None
) -> typing.AsyncIterator[vertexai.language_models.TextGenerationResponse]

Asynchronously gets a streaming model response for a single prompt.

The result is a stream (generator) of partial responses.

Parameter
Name	Description
`prompt`	`str` Question to ask the model.

tune_model

tune_model(
 training_data: typing.Union[str, pandas.core.frame.DataFrame],
 *,
 train_steps: typing.Optional[int] = None,
 learning_rate_multiplier: typing.Optional[float] = None,
 tuning_job_location: typing.Optional[str] = None,
 tuned_model_location: typing.Optional[str] = None,
 model_display_name: typing.Optional[str] = None,
 tuning_evaluation_spec: typing.Optional[TuningEvaluationSpec] = None,
 accelerator_type: typing.Optional[typing.Literal["TPU", "GPU"]] = None,
 max_context_length: typing.Optional[str] = None
) -> _LanguageModelTuningJob

Tunes a model based on training data.

This method launches and returns an asynchronous model tuning job. Usage:

tuning_job = model.tune_model(...)
... do some other work
tuned_model = tuning_job.get_tuned_model() # Blocks until tuning is complete

Parameter
Name	Description
`training_data`	`typing.Union[str, pandas.core.frame.DataFrame]` A Pandas DataFrame or a URI pointing to data in JSON lines format. The dataset schema is model-specific. See https://cloud.google.com/vertex-ai/docs/generative-ai/models/tune-models#dataset_format

Exceptions
Type	Description
`ValueError`	If the "tuning_job_location" value is not supported
`ValueError`	If the "tuned_model_location" value is not supported
`RuntimeError`	If the model does not support tuning

tune_model_rlhf

tune_model_rlhf(
 *,
 prompt_data: typing.Union[str, pandas.core.frame.DataFrame],
 preference_data: typing.Union[str, pandas.core.frame.DataFrame],
 model_display_name: typing.Optional[str] = None,
 prompt_sequence_length: typing.Optional[int] = None,
 target_sequence_length: typing.Optional[int] = None,
 reward_model_learning_rate_multiplier: typing.Optional[float] = None,
 reinforcement_learning_rate_multiplier: typing.Optional[float] = None,
 reward_model_train_steps: typing.Optional[int] = None,
 reinforcement_learning_train_steps: typing.Optional[int] = None,
 kl_coeff: typing.Optional[float] = None,
 default_context: typing.Optional[str] = None,
 tuning_job_location: typing.Optional[str] = None,
 accelerator_type: typing.Optional[typing.Literal["TPU", "GPU"]] = None,
 tuning_evaluation_spec: typing.Optional[TuningEvaluationSpec] = None
) -> _LanguageModelTuningJob

Tunes a model using reinforcement learning from human feedback.