b10103

ggml-org/llama.cppb10103Jul 24, 2026by github-actions[bot]same day as b10099

What changed since b10099

low risk

b10103 adds f16 support to the Metal leaky ReLU kernel, a small backend-specific change over b10099's CUDA NVFP4 work.

New features

  • Metal backend's leaky ReLU activation now accepts f16 tensors, extending GPU coverage beyond the previously supported types.

Improvements

  • Metal kernel coverage broadened for f16 activation paths, which can reduce fallback to other code paths on Apple GPUs.

Upgrade notes

No configuration or model changes are required; this is a drop-in build bump. Apple Silicon/Metal users running models that use leaky ReLU activations may see slightly better GPU utilization. The CUDA NVFP4 quantization improvements from b10099 remain unchanged.

Analyzed Nov 1, 2026

Upgrading?

Release notes summary

Adds f16 type support to the leaky relu activation function in the Metal backend.

Key Highlights

  • Add f16 type support to leaky relu
  • Metal backend update

Release Notes (excerpt)

<details open> metal : add f16 type support to leaky relu (#25981) </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