v1.36.1

1.36.1 (2023-11-07)

Features

• Add per_crowding_attribute_neighbor_count, approx_num_neighbors, fraction_leaf_nodes_to_search_override, and return_full_datapoint to MatchingEngineIndexEndpoint find_neighbors (33c551e)
• Add profiler support to tensorboard uploader sdk (be1df7f)
• Add support for per_crowding_attribute_num_neighbors approx_num_neighborsto MatchingEngineIndexEndpoint match() (e5c20c3)
• Add support for per_crowding_attribute_num_neighbors approx_num_neighborsto MatchingEngineIndexEndpoint match() (53d31b5)
• Add support for per_crowding_attribute_num_neighbors approx_num_neighborsto MatchingEngineIndexEndpoint match() (4e357d5)
• Enable grounding to ChatModel send_message and send_message_async methods (d4667f2)
• Enable grounding to TextGenerationModel predict and predict_async methods (b0b4e6b)
• LLM - Added support for the enable_checkpoint_selection tuning evaluation parameter (eaf4420)
• LLM - Added tuning support for the *-bison-32k models (9eba18f)
• LLM - Released CodeChatModel tuning to GA (621af52)

Bug Fixes

• Correct class name in system test (b822b57)

Documentation

• Clean up RoV create_ray_cluster docstring (1473e19)

Miscellaneous Chores

• Release 1.36.1 (1cde170)

v1.36.0

1.36.0 (2023-10-31)

Features

• Add preview count_tokens method to CodeGenerationModel (96e7f7d)
• Allow the users to use extra serialization arguments for objects. (ffbd872)
• Also support unhashable objects to be serialized with extra args (77a741e)
• LLM - Added count_tokens support to ChatModel (preview) (01989b1)
• LLM - Added new regions for tuning and tuned model inference (3d43497)
• LLM - Added support for async streaming (760a025)
• LLM - Added support for multiple response candidates in code chat models (598d57d)
• LLM - Added support for multiple response candidates in code generation models (0c371a4)
• LLM - Enable tuning eval TensorBoard without evaluation data (eaf5d81)
• LLM - Released CodeGenerationModel tuning to GA (87dfe40)
• LLM - Support accelerator_type in tuning (98ab2f9)
• Support experiment autologging when using persistent cluster as executor (c19b6c3)
• Upgrade BigQuery Datasource to use write() interface (7944348)

Bug Fixes

• Adding setuptools to dependencies for Python 3.12 and above. (afd540d)
• Fix Bigframes tensorflow serializer dependencies (b4cdb05)
• LLM - Fixed the async streaming (41bfcb6)
• LLM - Make tuning use the global staging bucket if specified (d9ced10)
• LVM - Fixed negative prompt in ImageGenerationModel (cbe3a0d)
• Made the Endpoint prediction client initialization lazy (eb6071f)
• Make sure PipelineRuntimeConfigBuilder is created with the right arguments (ad19838)
• Make sure the models list is populated before indexing (f1659e8)
• Raise exception for RoV BQ Write for too many rate limit exceeded (7e09529)
• Rollback BigQuery Datasource to use do_write() interface (dc1b82a)

v1.35.0

1.35.0 (2023-10-10)

Features

• Add serializer.register_custom_command() (639cf10)
• Install Bigframes sklearn dependencies automatically (7aaffe5)
• Install Bigframes tensorflow dependencies automatically (e58689b)
• Install Bigframes torch dependencies automatically (1d65347)
• LLM - Added support for multiple chat response candidates (587df74)
• LLM - Added support for multiple text generation response candidates (c3ae475)

Bug Fixes

• Duplicate logs in Colab (9b75259)
• LLM - Fixed tuning and evaluation when explicit credentials are specified (188dffe)
• Resolve Artifact Registry tags when creating PipelineJob (f04ca35)
• Resolve Artifact Registry tags when creating PipelineJob (06bf487)

Documentation

• Add probabilistic inference to TiDE and L2L model code samples. (efe88f9)

v1.34.0

1.34.0 (2023-10-02)

Features

• Add Model Garden support to vertexai.preview.from_pretrained (f978200)
• Enable vertexai preview persistent cluster executor (0ae969d)
• LLM - Added the count_tokens method to the preview TextGenerationModel and TextEmbeddingModel classes (6a2f2aa)
• LLM - Improved representation for blocked responses (222f222)
• LLM - Released ChatModel tuning to GA (7d667f9)

Bug Fixes

• Create PipelineJobSchedule in same project and location as associated PipelineJob by default (c22220e)

Documentation

• Add documentation for the preview namespace (69a67f2)

v1.33.1

1.33.1 (2023-09-20)

Bug Fixes

• Lightning trainer fails to be unwrapped in remote training (8271301)

v1.33.0

1.33.0 (2023-09-18)

Features

• Add Custom Job support to from_pretrained (8b0add1)
• Added async prediction and explanation support to the Endpoint class (e9eb159)
• LLM - Added support for async prediction methods (c9c9f10)
• LLM - CodeChat - Added support for context (f7feeca)
• Release Ray on Vertex SDK Preview (3be36e6)

Bug Fixes

• Handle Ray image parsing error (41a3a83)
• Vizier - Fixed field existence checks for child params in to_proto(). (d516931)

v1.32.0

1.32.0 (2023-09-05)

Features

• LLM - Added stop_sequences parameter to streaming methods and CodeChatModel (d62bb1b)
• LLM - Improved the handling of temperature and top_p in streaming (6566529)
• Support bigframes sharded parquet ingestion at remote deserialization (Tensorflow) (a8f85ec)
• Release Vertex SDK Preview (c60b9ca)
• Allow setting default service account (d11b8e6)

Bug Fixes

• Fix feature update since no LRO is created (468e6e7)
• LLM - CodeGenerationModel now supports safety attributes (c2c8a5e)
• LLM - Fixed batch prediction on tuned models (2a08535)
• LLM - Fixed the handling of the TextEmbeddingInput.task_type parameter. (2e3090b)
• Make statistics Optional for TextEmbedding. (7eaa1d4)

v1.31.1

1.31.1 (2023-08-24)

• fix: LLM - De-hardcoded the max_output_tokens default value for the CodeGenerationModel (f5a20eb)

v1.31.0

1.31.0 (2023-08-21)

Features

• Add disable_retries option to custom jobs. (db518b0)
• LLM - Added support for stop_sequences in inference (6f7ea84)
• LLM - Exposed the TextGenerationResponse.raw_prediction_response (f8f2b9c)
• LLM - Made tuning asynchronous when tuning becomes GA (226ab8b)
• LLM - release model evaluation for TextGenerationModel to public preview (8df5185)
• LLM - Released TextGenerationModel tuning to GA (62ff30d)
• LLM - Support streaming prediction for chat models (ce60cf7)
• LLM - Support streaming prediction for code chat models (0359f1d)
• LLM - Support streaming prediction for code generation models (3a8348b)
• LLM - Support streaming prediction for text generation models (fb527f3)
• LLM - TextEmbeddingModel - Added support for structural inputs (TextEmbeddingInput), auto_truncate parameter and result statistics (cbf9b6e)
• LVM - Added support for Image Generation models (b3729c1)
• LVM - Released ImageCaptioningModel to GA (7575046)
• LVM - Released ImageQnAModel to GA (fd5cb02)
• LVM - Released MultiModalEmbeddingModel to GA (e99f366)
• LVM - Removed the width and height parameters from ImageGenerationModel.generate_images since the service has dropped support for image sizes and aspect ratios (52897e6)
• Scheduled pipelines client GA. (62b8b23)

Documentation

• Generate documentation for tune_model and related class (705e1ea)
• LVM - Added autogenerated documentation for visual models (18e8bb2)

v1.30.1

1.30.1 (2023-08-11)

Features

• LLM - Added tuning support for chat-bison models (3a97c52)
• LLM - Added tuning support for codechat-bison models (af6e455)

Bug Fixes

• LLM - Fixed the TextGenerationModel.predict parameters (f3b25ab)

Miscellaneous Chores

• Release 1.30.1 (d1c79c4)