TL;DR: The latest llama.cpp release fixes context layer comparison issues and unifies the tool parser for LFM2/LFM2.5 models.
Summary: Release b9537 of llama.cpp introduces a fix for off-by-one comparisons related to gpu layers. Additionally, it unifies and fixes the tool parser within the common chat utilities for LFM2 and LFM2.5 models. The update also includes minor variable renaming for model layers across various supported platforms.
Why it matters: This release stabilizes gpu-offloading configurations and improves function calling reliability for supported models. Developers utilizing llama.cpp for local inference should update to avoid layer allocation errors and tool parsing bugs.
Source: github.com