Class TextGenerationModel (1.46.0)

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[int, float]] = 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[int, float]] = 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[int, float]] = 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[int, float]] = 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.

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

tuning_job = model.tune_model_rlhf(...)
... do some other work
tuned_model = tuning_job.get_tuned_model()  # Blocks until tuning is complete
Exceptions
Type Description
ValueError If the "tuning_job_location" value is not supported
RuntimeError If the model does not support tuning