The llama.cpp project’s b12099 release adds initial support for the Lightning Indexer operation on Hexagon hardware. The changes include HMX acceleration paths for prefill and decode, plus support for q8_0 decode gathering and q4_0 input. The release page lists Linux and Android builds for Snapdragon devices using the CPU, Adreno GPU and Hexagon NPU; it also links to setup guides.
Why it matters
Developers using llama.cpp on Snapdragon devices can try its new Hexagon path, though the release notes give no performance figures or model compatibility details.
