b10203

ggml-org/llama.cppb10203Jul 31, 2026by github-actions[bot]same day as b10202

What changed since b10202

low risk

SYCL backend gains Q2_0 quantized mat-mul support, one build after an unrelated SYCL kernel-fusion optimization.

New features

  • SYCL backend can now execute mul_mat operations on Q2_0-quantized tensors.
  • Additional Q2_0 layout/shape cases are handled beyond the initial implementation.

Improvements

  • Broadens the set of quantization formats usable with SYCL acceleration, reducing fallback to non-quantized paths.

Upgrade notes

Drop-in replacement for b10202 with no config or API changes; only SYCL users running Q2_0 models see behavioral differences. Rebuild or re-download the SYCL-enabled binary to pick up the new kernel. Non-SYCL backends are unaffected by this delta.

Analyzed Oct 31, 2026

Upgrading?

Release notes summary

This release adds support for the Q2 quantization format in matrix multiplication operations within the SYCL backend.

Key Highlights

  • Added support for Q2_0 quantization in `mul_mat` operations in the SYCL backend
  • Expanded support for various Q2_0 cases

Release Notes (excerpt)

<details open> [SYCL] Support q2 mul_mat (#26231) * support q2_0 in mul_mat * support more q2_0 case </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) - Ubuntu x64 (Vulkan) - Ubuntu arm64 (Vulkan) - Ubuntu x64 (ROCm 7.2) - Ubuntu x64 (OpenVINO) - Ubuntu x64 (SYCL FP32)…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories