llama.cpp b12322 Pre-release, Including Graph Memory, SYCL and CUDA Fixes
Posted As of
The b12322 pre-release was published on November 12 at 18:59 UTC, and consecutive builds from b12308 through b12322 were posted the same day. The included changes are the fixes listed on the release page.
What happened
- Based on the ggml-org/llama.cpp GitHub release page
- The b12322 pre-release was published on November 12 at 18:59 UTC.
- A fix to avoid touching the unused graph metadata page (#32043) is included.
- A fix to handle cases over 4GB in SYCL AOT (#31873) is included.
- A change to skip empty tiles in CUDA MoE MMQ stream-k (#32067) is included.
- Addition of decomposed FA for Adreno (#32046) is included.
Our take
If you build and use llama.cpp directly on a Mac or PC, it is worth checking this build's list of changes. However, the performance figures, benchmarks, and list of supported models for each PR have not been confirmed, so how much faster it actually gets cannot be known.
Sources and related posts
- ggml-org/llama.cpp GitHub Releases Primary source