b10106

ggml-org/llama.cppb10106Jul 24, 2026by github-actions[bot]same day as b10105

What changed since b10105

low risk

b10106 is a narrow CUDA build fix for q1_0 MMQ external compilation, with no functional or API changes over b10105.

Improvements

  • CUDA q1_0 MMQ kernels can now be compiled outside the main build tree, unblocking external/out-of-tree builds.

Bug fixes

  • Fixed a compilation failure in the q1_0 Multi-Quantization Matrix CUDA kernels when built externally.

Upgrade notes

This is a drop-in patch over b10105 with no config or CLI changes; only rebuild if you compile CUDA kernels externally or use q1_0 quantization. If you are jumping from a build older than b10105, note that mmap/mlock/directio flags were already replaced by the unified `load-mode` argument in that release.

Analyzed Nov 1, 2026

Upgrading?

Release notes summary

Fixes an issue preventing external compilation of the q1_0 Multi-Quantization Matrix (MMQ) kernels.

Key Highlights

  • Fix external compilation of q1_0 MMQ
  • CUDA kernel fix

New Features

  • Fix for external compilation of q1_0 MMQ

Release Notes (excerpt)

<details open> CUDA: fix external compilation of q1_0 MMQ (#25778) </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