Back to Mistral Rs

mistralrs-flash-attn

mistralrs-flash-attn/README.md

0.8.6201 B
Original Source

mistralrs-flash-attn

Fork of Candle's FA2 CUDA kernels for mistral.rs-owned attention paths.

Original source: candle-flash-attn from Hugging Face Candle, with custom modifications for mistral.rs.