The llama.cpp project has published pre-release b12173, adding OpenCL support for repeat operations with f16, i32, i16 and bf16 data types. OpenCL is a way to run computing workloads on supported processors and graphics hardware.
The release page lists builds for platforms including Linux, Windows, macOS, iOS and Android, with several hardware and accelerator options. It does not specify which devices or workloads benefit most from the change.
Why it matters
Developers using llama.cpp with OpenCL may be able to run repeat operations on these four data types, though the release notes do not say which devices gain support in practice.
