b9549

ggml-org/llama.cppb9549Jun 7, 2026by github-actions[bot]same day as b9548

What changed since b9548

medium risk

b9549 adds Gemma4 Multi-Token Prediction support on top of b9548's vocabulary compatibility fix.

New features

  • Gemma4 models can now use Multi-Token Prediction (MTP) for faster, more efficient generation.

Improvements

  • Build artifacts refreshed across all platforms (macOS, Linux CPU/Vulkan/ROCm/OpenVINO, iOS) for the new revision.

Upgrade notes

No configuration or migration steps are required; the vocabulary compatibility fix from b9548 is already carried forward. MTP for Gemma4 is an added capability, so existing Gemma4 workflows continue to work unchanged unless you explicitly enable it. Rebuild or re-pull binaries only if you want the new MTP path.

Analyzed Oct 31, 2026

Upgrading?

Release notes summary

This release adds support for Gemma4 Multi-Token Prediction (MTP) capability to enable more efficient text generation.

Key Highlights

  • Added Gemma4 Multi-Token Prediction (MTP) support.
  • New model capability for improved text generation.

Release Notes (excerpt)

<details open> llama : add Gemma4 MTP (#23398) </details> **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) DISABLED **Android:** - Android arm64 (CPU) **Windows:** - Windows x64 (CPU) - Windows arm64 (CPU)…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories