v1.52.0

1.52.0 (2024-05-21)

Features

• Add FeatureGroup delete (f9011e0)
• Add support for ToolConfig in the LangChain template (9bda328)
• Create Vertex Experiment when uploading Tensorboard logs (339f8b6)
• GenAI - Add BatchPredictionJob for GenAI models (df4a4f2)
• GenAI - Add cancel, delete, list methods in BatchPredictionJob (7ff8071)
• GenAI - Added the BatchPredictionJob.submit method (4d091c6)
• Private Endpoints - Added private service connect support to prediction endpoint. (6bdcfb3)

Bug Fixes

• Add validation for evaluation dataset fields, update logging info for eval api request count (d6ef500)
• Fix feature attribution drift visualization for model monitoring SDK (710f33d)
• Fix the default value of response_column_name in EvalTask.evaluate() (98f9b35)
• Update get_experiment_df to pass Experiment and allow empty metrics. (de5d0f3)

Documentation

• Add Vertex Model Monitoring V2 SDK documentation (b47e6ff)
• Update docstrings for rapid evaluation library. (d6d371d)

v1.51.0

1.51.0 (2024-05-10)

Features

• Add FeatureGroup create function (3938107)
• Add FeatureGroup init/get (e47d436)
• Add support for BaseModels in LangChain templates (5eb885e)
• Added the vision_models.Image._mime_type property to make vision_models.Image compatible with generative_models.Image (6557d88)
• AutoSxS Pairwise Metric in Rapid Evaluation SDK (b0c5eda)
• GenAI - Grounding - Released Google Web Search retriever to GA (32b030a)
• GenAI - Tuning - Supervised - Added support for the adapter_size parameter (88188d2)
• LLM - Made the tuning location parameters truly optional (bae8429)
• LLM - Support tuning of new text embedding models by migrating to the new v1.1.3 pipeline. (7fea754)
• LLM - Text embedding - Added the output_dimensionality and learning_rate_multiplier parameters to text embedding tuning (Preview only) (cc8bc96)
• LLM - Text Embedding - Added validation for text embedding tuning parameters. (5a300c1)
• Release Ray on Vertex SDK to GA (f7c5132)
• Support custom service account for Ray cluster creation and Ray Client connection (e0c6227)
• Support vector_distance_threshold filtering and file-based retrieval for RAG (cd85d8f)

Bug Fixes

• A bug in the evaluation library where the job crashes if only custom metrics are specified. (c528b6f)
• Add DeprecationWarning to vertexai.preview predictive models SDK (3c3727b)
• Add MAX_TOKENS to the list of successful finish reasons for Rapid Evaluation SDK (195c77e)
• AttributeError for TorchModelSerializer.deserialize in torch >=2.3.0 (20b1866)
• GenAI - Fixed handling of multiple tools in AutomaticFunctionCallingResponder (58e6ac9)
• Remove InternalServerError and Unknown evaluation service error from retriable exceptions (12c147b)
• Upload the reference model in model registry (510c833)

v1.50.0

1.50.0 (2024-05-02)

Features

• Add Candidate.grounding_metadata property (b22a8b8)
• Add option to not include time_series_metrics in get_experiment_df call. This will improve execution time for Experiments with large number of runs. (78a95c5)
• Add tune_model and deploy_tuned_model for TextEmbeddingModel. (42f5d6f)
• Automatically populate parents for full resource name in Vertex RAG SDK (26657ff)
• Deploy a tuned text embedding model -- it doesn't matter, if it's tuned using Node.js, or curl. (8ca9cdf)
• Make get_embeddings work both for foundational & tuned models. (b8b589c)
• Python SDK for Vertex Model Monitoring V2. (021d59f)
• Support public endpoint for Ray Client (57a5f78)

Bug Fixes

• Add deprecation warnings when using Ray v2.4 (3a36784)
• Append allowed_plugins in tb-gcp-uploader to default allowed plugins (aab9c3e)
• LLM - Added missing parameters to the no-op _TunableTextEmbeddingModelMixin.get_tuned_model method (eb05ac4)
• LVM - Fixed the typo in the VisionModel aspect ratio type annotation (2d19137)
• Move torch import (e6d34df)
• Ray - Fixed exception when using Ray 2.4 (2661f52)

v1.49.0

1.49.0 (2024-04-27)

Features

• Add additional parameters for GenerationConfig (0599ca1)
• Add FeatureNormType to MatchingEngineIndexConfig. (c0e7acc)
• Add mappings to pipeline templates for text-embedding models. (6279924)
• Add support for TPU v5 lite pod(v5e) for custom training jobs. Custom training jobs now accept the v5e machine types as listed in https://cloud.google.com/tpu/docs/tpus-in-gke#v5e. (415912e)
• Fix typo in Vertex Feature Store SDK. (b5404e7)
• GenAI - Improved the exception messages when candidates, parts or text are not available (e82264d)
• Support PreflightValidation in Preview PipelineJob submit function. (e88dc0d)
• Tune_model method for class TextEmbeddingModel. (3eda55d)
• Vertex RAG for enhanced generative AI (39b5149)

Bug Fixes

• GCS Bucket subdirectory not being creating while using ReasoningEngine create method (3d22a18)
• GenAI - Fixed the Part.mime_type property (819a44e)
• Handle missing import for RunnableConfig when generating schema for LangChain templates (76c5d6d)
• Parse intermediate steps from LangChain into JSON. (754c89d)
• Register TensorFlow models from Ray checkpoints for more recent TensorFlow version, addressing the deprecation of SavedModel format in keras 3 (1341e2c)
• Tensorboard uploader - move remaining tensorboard uploader code to aiplatform (f34094b)

v1.48.0

1.48.0 (2024-04-17)

Features

• Add support for reading requirements from a file. (80db7a0)
• Adding tpu_topology to Vertex SDK (423c764)
• Enable continuous upload for profile logs. (f05924d)
• GenAI - Added the GenerationResponse.prompt_feedback property (efd5a72)
• GenAI - Added the GenerationResponse.usage_metadata property (0654c35)
• Support NOT_EQUAL for MatchingEngineIndexEndpoint numeric_restricts. (aa918e3)
• Support referenced models in SDK. (c9b6b8b)

Bug Fixes

• Add validation check for extra_packages when creating a reasoning engine. (255dabc)
• Add validation for langchain tools. (a821d50)
• Fixed the vertexai.init partial initialization issues (636a654)
• GenAI - Workaround for streaming when content role is missing in service responses (fa35b91)

Documentation

• Add Reasoning Engine reference documentation (496fc4b)
• GenAI - Add Rapid Evaluation SDK reference documentation (40b728b)

v1.47.0

1.47.0 (2024-04-06)

Features

• Add display experiment button for tuning in Ipython environments (9bb687c)
• Add Persistent Resource ID parameter to Custom Job form_local_script, run, and submit methods. (f5be0b5)
• Add Persistent Resource Id parameter to Custom Training Job run and submit methods. (f428006)
• Added GA support for PersistentResource management (98a07dd)
• Added GA support for running Custom and Hp tuning jobs on Persistent Resources (35ecbac)
• Added the Experiment.dashboard_url property (c8eec21)
• GenAI - Added support for SafetySetting.method (probability or severity) (317ab8f)
• GenAI - Added support for supervised fine-tuning (036d2d0)
• GenAI - Added support for system instructions (4990eb6)
• GenAI - Forced function calling feature (806ef9f)
• Initial template for Langchain on Vertex. (0752a29)
• LLM - Add RLHF-tuning support for text-bison@002 (1f27c3e)
• Vertex AI Extension SDK Public Preview (137b5e1)
• Vertex AI Reasoning Engine SDK Public Preview (6aaa5d0)
• Vertex Rapid Evaluation SDK and Prompt Template for Vertex Prompt Management Public Preview (8c6ddf5)

Bug Fixes

• GenAI - Fixed response validation error during streaming (c881998)
• GenAI - Fixed the GenerativeModel's handling of tuned models from different region (bf33fb3)
• GenAI - Fixed the TuningJob dashboard URL (5367fbb)
• Reinstate persistent resource preview class and tests (765d60d)

Documentation

• Add run custom job on persistent resource sample. (53fc845)
• Add run custom job on persistent resource sample. (31100c6)

v1.46.0

1.46.0 (2024-03-30)

Features

• GenAI - Automatic Function Calling feature (eef84c6)
• Python SDK for Vertex Feature Store. (5015d25)

v1.45.0

1.45.0 (2024-03-28)

Features

• Add an arg to turn off Ray metrics collection during cluster creation (e33d11f)
• Add Batch Delete and Batch Cancel Pipeline Jobs and unit tests. (cb495e7)
• Add display model evaluation button for Ipython environments (181dc7a)
• Add safety filter levels, watermark support and person generation support for Imagen 2 (0c498c5)
• Add safety filter levels, watermark support and person generation support for Imagen 2 (e2efdbe)
• Add support for output_dimensionality parameter through get_embeddings. (b1cab3f)
• Add support for task types (TASK_QUESTION_ANSWERING, TASK_FACT_VERIFICATION) through get_embeddings. (9ec4590)
• Add v1beta1 service clients for reasoning engine (13ec7e0)
• GenAI - Support generating JSON Schema from Python function (be4922a)
• Improve get_experiment_df execution speed (2e56acc)
• Ray on Vertex enables XGBoost register model with custom version using pre-built container (e45ef96)
• Support accelerator_type in RLHF tuning (e51c977)

Bug Fixes

• Fix failing get_experiments_df test. (767712e)
• GenAI - Capture content blocked case when validating responses (f0086df)
• LLM - Fixed the InlineContext grounding source (6f59100)
• Made Endpoint.raw_predict thread-safe by setting authorized_session last (c72c1ef)

v1.44.0

1.44.0 (2024-03-13)

Features

• Add custom tool context manager for telemetry (b30f5a6)
• Add display experiment button for Ipython environments (0b33bdd)
• Allow module_name to be specified with the base logger. (c01df66)
• Enable Ray cluster creation with custom_image for each Resource (f90c9cc)
• GenAI - Added function_calls shortcut property to Candidate class. (d6490ff)
• LLM - Added support for the max_context_length tuning parameter (e5daae9)
• Support shard_size for MatchingEngineIndex create index. (6dbf7d3)
• Use colab enterprise enviroment variables to infer project_id and region (5baf5f8)

Bug Fixes

• Batch read tensorboard time series data to allow reading more than 20 time series metrics. (6e3eac2)
• GenAI - Fixed get_tuned_model deploying the model every time (e1c7870)
• GenAI - Improved from_dict methods for content types (GenerationResponse, Candidate, Content, Part) (613ce69)
• Improve import time by moving TensorFlow to lazy import (f294ba8)
• LVM - Added support for GCS storage.googleapis.com URL import in vision_models.Image (2690e72)
• Raise AttributeError for accessing non-existent fields of a proto. (b91edf5)

Documentation

• Fixed the documentation generation (63ad1bf)
• GenAI - Added the GA classes to the documentation (9eb5a52)

v1.43.0

1.43.0 (2024-02-29)

Features

• Add update_mask to MatchingEngineIndex upsert_datapoints() to support dynamic metadata update. (81f6a25)
• Add Ray on Vertex BigQuery read/write support for Ray 2.9 (e048e3a)
• Add read_bigquery and write_bigquery wrappers for Ray on Vertex (4739118)
• Add transport override to enable the use of REST instead of GRPC (6ab4084)
• Enable Ray cluster creation and registering TensorFlow checkpoint to Vertex with Ray version 2.9 (ff148cd)
• LLM - Add support for batch prediction to CodeGenerationModel (code-bison) (fbf2f7c)
• Support custom timeout for MatchingEngineIndex and MatchingEngineIndexEndpoint APIs. (09d1946)
• Update ray extras installation in setup.py for Ray 2.9.3 (d947304)

Bug Fixes

• Allow destination directory to exist before we copy. (0b55762)
• LVM - Fixed VideoSegmentConfig in Multimodal Embeddings API call (665e78c)

Documentation

• Update transport docstrings to include that REST is in preview (a442866)