The llama.cpp project released a new version that adds Q1_0 and Q2_0 quantization for Qualcomm's Hexagon NPU. The update also tweaks tiling and vector dot operations to cut overhead. It works on Linux and Windows builds that target Snapdragon CPUs, Adreno GPUs and Hexagon NPUs, as well as other platforms. The changes are documented in the developer guide and signed in the release attestations.
Why it matters
Developers can now run larger language models on Snapdragon‑based phones and tablets with lower latency and power use.
