b8513
Latest update patches critical variable assertion error while adding new Windows and Linux build targets.
The open-source community behind llama.cpp, the high-performance C++ inference engine for Llama models, has released a new update identified as commit b8513. This release primarily addresses a bug in the SYCL (Intel oneAPI) backend where an incorrect variable was being checked by an assertion, potentially causing crashes or incorrect behavior when running models on Intel GPUs and accelerators. The fix ensures more stable execution for developers leveraging Intel's heterogeneous computing platform.
Alongside the critical bug fix, the release expands the project's extensive list of pre-built binaries. Notable new additions include a Windows x64 build with CUDA 13.1 DLLs for users on the latest NVIDIA toolkit, and several specialized builds for the openEuler Linux distribution. These openEuler targets support Huawei's Ascend AI processors (310P and 910B) using the ACL (Ascend Computing Language) Graph, significantly broadening the hardware ecosystem where llama.cpp can be deployed efficiently out-of-the-box.
- Fixes a SYCL backend bug where a wrong variable was checked by an assert statement (#20903).
- Adds new pre-built binary for Windows x64 with CUDA 13.1 DLLs, updating NVIDIA toolkit support.
- Expands openEuler Linux support with four new builds targeting Huawei Ascend 310P and 910B hardware.
Why It Matters
Ensures stability for Intel GPU AI workloads and expands accessible deployment options for enterprise and edge hardware.