The ggml-org team released llama.cpp version b12124 with a change to its OpenCL support: it can now handle sigmoid operations on non-contiguous data. OpenCL is a framework for running computing tasks on different processors.
The release page also lists prebuilt downloads for macOS, iOS and Ubuntu, including CPU and Vulkan builds. It does not describe broader performance changes or say which users will notice a difference.
Why it matters
Developers using llama.cpp with OpenCL can access the newly supported operation, but the release notes do not establish whether it improves speed or affects everyday users.
