b8981
Latest update fixes a key sampler bug and adds builds for more platforms...
Deep Dive
llama.cpp released version b8981. The update includes a fix for the reasoning budget sampler (issue #22488) and adds pre-built binaries for Windows arm64 (CPU), Android arm64 (CPU), and openEuler (x86 and aarch64) with ACL Graph support.
Key Points
- Critical bug fix: prompt tokens no longer passed to reasoning budget sampler (issue #22488), improving inference accuracy
- New pre-built binaries for Windows arm64 (CPU), Windows x64 (HIP), and Android arm64 (CPU)
- Expanded support for openEuler Linux: x86 and aarch64 with 310p and 910b ACL Graph builds
Why It Matters
Broader platform support and a key bug fix make local AI inference more reliable and accessible.