b9918

ggml-org/llama.cppb9918Jul 8, 2026by github-actions[bot]same day as b9917

What changed since b9917

low risk

b9918 only adds Metal f16->f16 GGML_OP_SET_ROWS support on top of b9917's tokenizer security fix.

New features

  • Metal backend now supports GGML_OP_SET_ROWS when the source tensor is f16.

Improvements

  • Apple Silicon GPU path can now execute f16 set_rows operations natively instead of falling back or failing.

Upgrade notes

The delta between these two builds is limited to a Metal kernel addition; the UGM tokenizer bounds-check hardening was already present in b9917. No config, model, or API changes are required, and non-Metal platforms see no behavioral difference.

Analyzed Oct 30, 2026

Upgrading?

Release notes summary

Adds support for f16->f16 GGML_OP_SET_ROWS operation on Metal (Apple Silicon).

Key Highlights

  • Metal support for f16->f16 GGML_OP_SET_ROWS

Release Notes (excerpt)

<details open> metal : add set_rows with src0 f16 (#25434) Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> </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:**…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories