b8798

The llama.cpp project, a leading C++ implementation for running Meta's Llama models efficiently, has pushed a significant new update tagged as b8798. This release primarily addresses a technical bug (issue #21939) where the configured context window size parameter (n_ctx) was not being properly retrieved after initializing a model's context. This fix ensures stable performance and correct memory allocation for long-context interactions, a critical feature for applications using retrieval-augmented generation (RAG) or document analysis.

Beyond the bug fix, the update significantly broadens the project's hardware compatibility. The release now includes pre-compiled binaries for new backends, notably adding Windows support for the HIP runtime (commonly used for AMD GPUs) and expanding its reach on the openEuler OS with builds for Huawei's Ascend AI processors (310p and 910b). This continues llama.cpp's mission of democratizing local AI by supporting an ever-wider array of consumer and server-grade hardware, from Apple Silicon and NVIDIA CUDA to Intel OpenVINO and now AMD HIP.