b10305
ggml-org/llama.cppb10305Aug 7, 2026by github-actions[bot]same day as b10303
What changed since b10303
b10305 adds SYCL backend support for DSv4 ops (LIGHTNING_INDEXER and DSV4_HC variants) plus an ops.md doc refresh.
New features
- SYCL backend now implements the LIGHTNING_INDEXER operation, enabling that model path on Intel GPUs.
- SYCL backend gains the DSV4_HC_COMB, DSV4_HC_POST and DSV4_HC_PRE operations for DSv4 model support.
Improvements
- ops.md documentation updated to reflect the newly supported SYCL operations.
- Minor code formatting cleanup accompanying the SYCL op additions.
Upgrade notes
This is an additive, backend-specific change with no CLI, config, or model-format changes; swapping in the new binaries is sufficient. Only SYCL users running DSv4-style models will see behavioral differences. The Arc770 FLASH_ATTN_EXT fix from b10303 remains in place.
Analyzed Oct 30, 2026
Upgrading?
Release notes summary
This release adds support for specific DSv4 (Data Shape v4) operations in the SYCL backend, including LIGHTNING_INDEXER and various DSV4_HC variants.
Key Highlights
- Added support for DSv4 operations: LIGHTNING_INDEXER, DSV4_HC_COMB, DSV4_HC_POST, DSV4_HC_PRE (#26568)
- Updated ops.md documentation
Release Notes (excerpt)
<details open> sycl : Support DSv4 OPs: LIGHTNING_INDEXER,DSV4_HC_COMB,DSV4_HC_POST,DSV4_HC_PRE (#26568) * support DSv4 OPs: LIGHTNING_INDEXER,DSV4_HC_COMB,DSV4_HC_POST,DSV4_HC_PREwq * update ops.md * fix format issue </details> **Website:** - <https://llama.app> **macOS/iOS:** - macOS Apple Silicon (arm64) - macOS Apple Silicon (arm64, KleidiAI enabled) DISABLED - macOS Intel (x64) - iOS XCFramework **Linux:** - Ubuntu x64 (CPU) - Ubuntu arm64 (CPU) - Ubuntu s390x (CPU)…
Published by ggml-org on GitHub.
Read the full release notesRelated Repositories
:electron: Build cross-platform desktop apps with JavaScript, HTML, and CSS
Godot Engine – Multi-platform 2D and 3D game engine
ClickHouse® is a real-time analytics database management system
Self-hosted game stream host for Moonlight.
DuckDB is a high-performance analytical database system