llama.cpp b11019 fixes embedded GGUF loading bugs

How it developed

1. September 18, 2026llama.cpp ReleasesWhere it started

   llama.cpp b11019 fixes embedded GGUF loading bugs

2. September 19, 2026llama.cpp Releases

   llama.cpp fixes CUDA argsort corruption bug

   llama.cpp fixes CUDA argsort corruption bug caused by memory aliasing in CUB radix sort during top-k selection

3. September 19, 2026llama.cpp Releases

   llama.cpp b11039 fixes SWA pattern roundtrip

   llama.cpp b11039 fixes SWA pattern roundtrip by explicitly writing per-layer SWA flags to ensure bit-exact conversions for models like OLMo2