b9693

ggml-org/llama.cppb9693Jun 18, 2026by github-actions[bot]same day as b9692

What changed since b9692

low risk

b9693 adds a BF16 capability check to the Metal concat kernel, with no other changes over b9692.

New features

  • Metal concat kernel now verifies BF16 support before executing, gating the path on hardware capability.

Improvements

  • Improved robustness of Metal backend on Apple GPUs that lack native BF16 support.

Bug fixes

  • Prevents incorrect or unsupported BF16 concat operations on Metal hardware without BF16 capability.

Upgrade notes

This is a single-commit increment over b9692 and requires no config or model changes. Apple Silicon users running BF16 models on Metal benefit most; other backends are unaffected. The LLaVA-UHD batch-dimension fix from b9692 remains included.

Analyzed Nov 1, 2026

Upgrading?

Release notes summary

Adds a specific check for BF16 support within the Metal concat kernel to ensure proper functionality.

Key Highlights

  • BF16 support verification in Metal concat kernel
  • Hardware compatibility check

Release Notes (excerpt)

<details open> metal : check for BF16 support in concat kernel (#24747) </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) - Ubuntu x64 (SYCL FP16) **Android:** - Android arm64 (CPU) **Windows:**…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories