Ray

This release introduces significant enhancements to Ray Serve LLM with KV cache and token-aware routing, and Ray Core with task event offloading. It also brings Databricks DeltaLake integration and a new shuffle v2 backend to Ray Data, alongside experimental Ray Sandbox and expanded TPU support across Ray Train and Core.

This release introduces significant improvements to Ray Data with DataSourceV2 enabled by default and Hash Shuffle V2. Ray Serve enhances its HAProxy ingress with a new PyPI package and gRPC support, while Ray Core adds an embedded RocksDB backend for GCS fault tolerance.

This release includes fixes for Ray Data's `to_pandas` regressions, early detection for system-slice memory pressure in Ray Core, and protobuf 7 compatibility with routing fixes for Ray Serve LLM direct streaming.

This release focuses heavily on Ray Data stability, introducing features like multi-dataset support, automatic batch sizing, and improved memory management. Ray Serve received significant LLM performance enhancements via re-architecture and new routing policies, while Ray Core introduced GPU-domain-aware placement groups and initial Kubernetes in-place pod resizing support.

This release focuses on infrastructure maintenance, fixing an SSH connectivity issue in the ray-llm image and upgrading apt packages in the slim base image.

This release introduces the new `DataSourceV2` API, GPU shuffle support, and significant enhancements to Ray Serve including gRPC streaming and HAProxy ingress. Numerous fixes address RCE vulnerabilities, scheduling races, and resource management issues across Ray Data.

This patch release fixes a critical performance degradation issue by temporarily disabling the hanging issue detector which was causing blocking calls to the Ray State API.

This release introduces significant new features in Ray Data, including checkpointing, expanded Compute Expressions, and Databricks UC credential support, alongside major enhancements to cluster autoscaling and performance optimizations. Ray Serve gains queue-based autoscaling for TaskConsumers and improved deployment observability.

This release adds major new features such as a utilization‑based autoscaler, Kafka datasource, and deployment topology visibility, while dropping Pydantic V1 support and removing deprecated APIs.

This release addresses CVE-2025-62593 by rejecting Sec-Fetch-* and other browser-specific headers in the dashboard's browser rejection logic.

Security update adds more robust handling for CVE-2025-62593 by checking extra browser-specific headers in dashboard rejection logic.

Ray 2.x introduces token authentication, extensive Iceberg enhancements, new expression capabilities, and numerous performance improvements, while dropping Python 3.9 support and adjusting several defaults that may require migration steps.

Added reuse of previous metadata for identical tensor list transfers using `nixl`.

This release introduces Ray Train v2 as the default, adds application‑level autoscaling to Ray Serve, and brings numerous new features and fixes to Ray Data, Train, Tune, and Serve.

Fixed a deadlock that occurred when cancelling stale requests on in-order actors.

This release adds major enhancements across Ray Data, Core, Train, Serve, and RLLib, including a new hash‑based shuffle, expression API, multi‑node LLM support, Direct Transport for GPU data, async inference in Serve, and extensive performance and memory optimizations.

There are no functional changes between version 2.49.2 and 2.49.1; this release only increments the patch version for out-of-band reasons.

This release addresses two bugs: GPU metrics missing in the Ray Dashboard and a regression in Ray Data handling large schemas.

This release adds extensive performance enhancements, new features such as JaxTrainer, async inference, and custom autoscaling, while fixing numerous bugs; it also introduces a few breaking changes that may require migration steps.

This release adds Delta Lake support, new collective operations, custom request routing, and offline policy evaluation, while also delivering numerous performance improvements and bug fixes across Ray Data, Train, Serve, Tune, and RLlib.

Ray 2.47.1 resolves a startup failure on macOS.

This release adds major LLM serving capabilities, custom routing, and numerous performance and observability improvements across Ray Data, Train, Serve, RLlib, and the Dashboard, while also introducing several breaking changes such as non‑blocking pipelines, stricter Serve name validation, and default uv run integration.

Ray 2.46 adds hash‑shuffle support for Data, vLLM v1 compatibility for Serve LLM, a new Train Grafana dashboard, and numerous performance and bug‑fix improvements.

Ray 2.45 adds configurable object store fallback, new cgraph transport options, a ClickHouse sink and several dataset API enhancements, along with numerous stability fixes and an upgraded LightGBM version.

There are no functional changes between version 2.44.1 and 2.44.0; this release only increments the patch version for out-of-band reasons.