mirror of
https://github.com/ggml-org/llama.cpp.git
synced 2026-09-09 06:19:04 +02:00
517b7170e1
Very similar implementation to the flash-attention chunking, with similar benefits.