b9741

ggml-org/llama.cppb9741Jun 20, 2026by github-actions[bot]same day as b9740

What changed since b9740

low risk

b9741 refactors quantization metadata access to use LLM_KV, replacing the Windows test-argv fix from b9740.

Improvements

  • Quantization version and file type metadata are now read and written through the LLM_KV abstraction instead of ad-hoc lookups.
  • Metadata handling is more consistent across model loading and saving paths.

Bug fixes

  • The Windows test-args-parser flakiness addressed in b9740 (guarding the win32 utf8 argv override) remains in place and is carried forward.

Upgrade notes

This is an internal refactor with no CLI, config, or GGUF format changes expected. If you parse or write GGUF metadata directly, confirm your key names for quantization version and file type still match what LLM_KV emits. No migration steps are required for typical inference users.

Analyzed Nov 1, 2026

Upgrading?

Release notes summary

Refactors how quantization metadata is stored and accessed by using LLM_KV for quantization_version and file_type.

Key Highlights

  • Refactors quantization metadata to use LLM_KV
  • Improves consistency in metadata storage and retrieval

New Features

  • Metadata refactoring for quantization

Release Notes (excerpt)

<details open> llama : use LLM_KV for quantization_version & file_type (#24802) Signed-off-by: Adrien Gallouët <angt@huggingface.co> </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)…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories