b10486

ggml-org/llama.cppb10486Aug 18, 2026by github-actions[bot]same day as b10485

What changed since b10485

low risk

b10486 is a narrow follow-up that fixes LFM2 image tiling threshold handling in the mtmd multi-modal path, including a Windows build fix.

Improvements

  • Multi-modal test coverage was refactored alongside the tiling fix, tightening validation of mtmd behavior.
  • The tiling correction was made portable, with an explicit follow-up fix so it also builds and runs on Windows.

Bug fixes

  • LFM2 image tiling threshold was being computed/interpreted incorrectly in mtmd, affecting multi-modal inference for LFM2 vision models.
  • A platform-specific issue introduced by the tiling change was resolved for Windows builds.

Upgrade notes

This is a drop-in patch over b10485 with no config, CLI, or API changes; only users running LFM2 multi-modal (mtmd) inference are meaningfully affected. If you are on b10485 and use LFM2 vision inputs, upgrade to get correct tiling behavior. Note that the ggml core sync landed in b10485, not here, so the two releases are not cumulative in scope.

Analyzed Oct 24, 2026

Upgrading?

Release notes summary

Resolves a bug in the multi-modal model (mtmd) where the LFM2 image tiling threshold was incorrectly processed, ensuring compatibility across platforms.

Key Highlights

  • Fix LFM2 image tiling threshold
  • Refactor testing for multi-modal models
  • Fix on Windows platform

New Features

  • Bug fix for multi-modal image tiling in mtmd

Release Notes (excerpt)

<details open> mtmd: fix LFM2 image tiling threshold (#27057) * mtmd: fix LFM2 image tiling threshold * refactor testing * fix * fix on windows --------- Co-authored-by: Xuan Son Nguyen <son@huggingface.co> </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)…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories