>>106271845
Things get implemented as they become popular. Everyone forgot about SWA, but they use it without even knowing. Or disable it entirely to have fast regens.
But now we have this and nobody even mentioned it.
>server : add SWA checkpoints
>https://github.com/ggml-org/llama.cpp/pull/15293

>>106272254
There's this one too.
>llama : add Xiaomi Mimo (with proper MTP - multi token predict)
>https://github.com/ggml-org/llama.cpp/pull/13236