CYBER WATCH / EQUINOXE INTELLIGENCE
← All articles
b12069: opencl: add initial iq4_xs support (#31339)
llama.cpp Releases · 2026-11-03 14:00:41
This release adds initial support for the iq4_xs quantization format in OpenCL. It introduces several optimizations and kernels for Adreno GPUs, including AOS and SOA matrix‑vector operations, tiled matrix multiplication, no‑shuffle GEMM/GEMV, DP4A support, and decoding optimizations for narrow batches.Read original source ↗