b10531

ggml-org/llama.cppb10531Aug 21, 2026by github-actions[bot]same day as b10520

What changed since b10520

medium risk

b10531 reverts the tensor-split meta backend fix, undoing behavior introduced earlier while keeping b10520's ARM FP16 build fix.

Breaking changes

  • The tensor-split meta backend fix (#26502) is reverted, so setups relying on the corrected tensor-split behavior will see the older semantics again.

Improvements

  • Retains b10520's gating of __fp16 on __ARM_FP16_FORMAT_IEEE, keeping 32-bit ARM NEON and NEON+FMA builds compiling via the scalar fallback path.

Bug fixes

  • Reverts the tensor-split meta backend change to resolve issues it introduced, restoring the previously working behavior.

Upgrade notes

If you depend on the tensor-split meta backend behavior from builds between #26502 and this revert, re-test multi-device splitting after upgrading. Users on 32-bit ARM with NEON keep the compile fix from b10520; no configuration changes are required. Pin to a build containing #26502 only if that specific fix is essential and you can tolerate its reported issues.

Analyzed Oct 26, 2026

Upgrading?

Release notes summary

Reverts a previous fix for the tensor-split meta backend to address issues introduced by the change, restoring the previous behavior.

Key Highlights

  • Reverted tensor-split meta backend fixes
  • Restored previous behavior for tensor splitting

Release Notes (excerpt)

<details open> Revert "tensor-split meta backend fixes (#26502)" (#27433) This reverts commit d59d455fd8ea09e5a2e87ce2a9d668267ffb5ccd. </details> **Website:** - <https://llama.app> **Attestations:** - <https://github.com/ggml-org/llama.cpp/attestations/42012577> **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)…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories