The ggml-org/llama.cpp release b12020 is a build published on October 31, 2026. The change listed in this release record is the addition of FWHT support for OpenCL, and it also announces distribution files for macOS and iOS. The repository description presents the project's role as LLM inference implemented in C/C++. So this news is that OpenCL support for a specific operation has been added in a new release of the inference project. The basis for it is the public b12020 release record of ggml-org/llama.cpp.
No detailed description of the FWHT operation or the scope of the support is given in the provided facts. What can be confirmed is that the release record explicitly names OpenCL as the FWHT support target in b12020. That information alone does not make it possible to explain which devices or models can use this path, or how it differs from the existing approach. The announcement of distribution files for macOS and iOS is likewise a fact included in the release record; the composition of each file and its conditions of use are not given. This update should therefore be read as limited to the explicitly stated support item and the distribution announcement, rather than having its scope broadly interpreted.
The provided record contains no performance figures or comparison results for the FWHT support. As a result, there is not enough basis to judge what change this addition brings to inference speed or resource usage. What the release record confirms is FWHT support for OpenCL and the announcement of the macOS and iOS distribution files, not a measured performance improvement. To speak about the actual effect of the new support would require separate measurement or more specific usage conditions, but such information is not among the facts currently given. The addition of support itself is confirmed in this release, but it is an update whose magnitude of effect cannot be asserted.