llama.cpp b10107 fixes Windows crash on Hexagon DSP
New release patches op_poll crash and expands platform support.
Deep Dive
Key Points
- Fixes Windows crash when op_poll is enabled on Hexagon DSP
- Prebuilt binaries for 20+ platform/configurations including CUDA 13 and KleidiAI
- Supports local LLM inference on Qualcomm Hexagon, ROCm, Vulkan, OpenVINO, and more
Why It Matters
llama.cpp b10107 ensures stable local AI inference on Windows with Hexagon DSP, critical for edge AI developers.