b10205

ggml-org/llama.cppb10205Jul 31, 2026by github-actions[bot]same day as b10204

What changed since b10204

low risk

b10205 adds a ZenDNN group matmul path for mul_mat_id and tunes its fallback threshold by expert count.

New features

  • ZenDNN backend now uses a direct group matmul API to accelerate the mul_mat_id operation.

Improvements

  • MUL_MAT_ID fallback threshold in the ZenDNN backend is now scaled by expert count, improving dispatch decisions for MoE-style models.

Upgrade notes

This is a single-commit backend optimization with no API, config, or CLI changes, so a drop-in replacement of the binaries is sufficient. Only users running the ZenDNN backend on Intel hardware will see any behavioral difference.

Analyzed Nov 1, 2026

Upgrading?

Release notes summary

This release focuses on optimizing group matrix multiplication for the Zendnn backend, specifically targeting the mul_mat_id operation for improved efficiency.

Key Highlights

  • Zendnn group matmul direct API for mul_mat_id
  • Scale MUL_MAT_ID fallback threshold adjusted by expert count

Release Notes (excerpt)

<details open> ggml-zendnn : group matmul direct API for mul_mat_id (#25918) * ggml-zendnn : group matmul API for mul_mat_id * ggml-zendnn : scale MUL_MAT_ID fallback threshold by expert count </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)…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories