llama.cpp b11046 adds OpenCL flash attention

How it developed

1. September 19, 2026llama.cpp ReleasesWhere it started

   llama.cpp b11046 adds OpenCL flash attention

2. September 22, 2026llama.cpp Releases

   llama.cpp adds Intel Xe flash attention optimization

3. September 24, 2026llama.cpp Releases

   llama.cpp b11140 optimizes sparse attention on CUDA

   llama.cpp b11140 optimizes sparse flash attention prefill for NVIDIA GPUs, reducing batched sparse ops from 586 to 244 microseconds