mirror of
https://github.com/ggml-org/llama.cpp.git
synced 2026-06-30 08:10:20 +00:00
517b7170e1
Very similar implementation to the flash-attention chunking, with similar benefits.