Changes in HeatWave 9.3.2 (2025-06-24, General Availability)

Note

These release notes were created with the assistance of HeatWave GenAI.

• HeatWave AutoML

• HeatWave GenAI

• HeatWave Lakehouse

HeatWave AutoML

• HeatWave AutoML now supports drift detection for anomaly detection, enabling you to identify changes in data distribution. With this update, you can review the model metadata and enable the additional_details option with the ML_PREDICT_TABLE and ML_PREDICT_ROW routines to get valuable insights into data drift, enhancing the overall anomaly detection experience.

  For more information, see Data Drift Detection. (WL #16368)

HeatWave GenAI

• Content generation using HeatWave GenAI now supports speculative decoding, which enables faster response token generation and speeds up text generation if the target Large Language Model (LLM) supports speculative decoding.

  For more information, see ML_GENERATE, ML_GENERATE_TABLE, and @chat_options Parameters. (WL #16809)

• Document ingestion using Auto Parallel Load has been enhanced to support customised segmentation of text during vector store creation. This enhancement lets you have greater control over how text is segmented, enabling more precise and tailored text analysis capabilities.

  For more information, see Ingesting Files Using Auto Parallel Load. (WL #16683)

• HeatWave GenAI now supports automated discovery and listing of all available LLMs and embedding models in HeatWave. This enhancement lets you stay up-to-date on model availability, and enables you to make informed usage decisions especially in the case of OCI Generative AI Service models as the list of models available in HeatWave might change before a new version of MySQL is available.

  For more information, see Supported LLMs, Embedding Models, and Languages. (WL #16664)

HeatWave Lakehouse

• HeatWave Lakehouse now supports the direct loading of compressed files from Object Storage into tables, enhancing data ingestion capabilities and simplifying data pipelines. This update lets you load gzip, zip, and bzip2 compressed files that contain CSV and JSON data into HeatWave. This enhancement streamlines the process of working with compressed data, making it more efficient and cost effective for you to manage and analyze your data within HeatWave Lakehouse.

  For more information, see External Table Syntax. (WL #16650)

• HeatWave Lakehouse now supports loading vector columns from CSV and Parquet files, enabling you to bring your own custom vector embeddings into HeatWave and leverage HeatWave GenAI features. With this enhancement, you can optimize and manage semantic search capabilities for your datasets and integrate with the large ecosystem of GenAI capabilities.

  For more information, see Supported File Formats and Data Types. (WL #16563)