Mutable.ai — Dao-AILab/flash-attention
LicenseBSD 3-Clause "New" or "Revised"12kRepository104Software Versionu-0.0.1Basic• • •The flash-attention repository provides a highly optimized and efficient implementation of the Transformer Attention mechanism, known as "Flash Attention". This library is particularly useful for engineers building transformer-based models, as it offers significant performance improvements over standard PyTorch implementations.
The core of the repository is the …/flash_attn directory, which contains the C++ an...
Read more at wiki.mutable.ai