b10233

ggml-org/llama.cppb10233Aug 2, 2026by github-actions[bot]same day as b10232

What changed since b10232

low risk

b10233 is a single OpenCL bugfix that caps GLU workgroup size; the Metal DeepSeek V4 kernels from b10232 carry over unchanged.

Improvements

  • OpenCL GLU kernel now constrains its local workgroup size, avoiding oversized workgroup configurations on some devices.

Bug fixes

  • Prevents potential instability or incorrect execution in the OpenCL backend when the GLU operation was dispatched with large workgroups.

Upgrade notes

This is a drop-in patch bump with no API, config, or model-format changes; only OpenCL users are affected. If you run OpenCL builds on hardware that previously showed GLU-related crashes or wrong outputs, upgrading is worthwhile. Metal users see no functional change from b10232.

Analyzed Oct 31, 2026

Upgrading?

Release notes summary

Fixed a potential issue in the OpenCL backend by limiting the local workgroup size for the GLU operation to prevent instability.

Key Highlights

  • Limited local workgroup size for GLU operation in OpenCL.
  • Addresses potential stability issues with large workgroups.
  • Bugfix for OpenCL backend kernel execution.

New Features

  • OpenCL GLU workgroup size limit

Release Notes (excerpt)

<details open> opencl: limit local workgroup size for GLU operation (#26383) </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) - Ubuntu x64 (SYCL FP16) **Android:**…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories