tspeterkim / flash-attention-minimal

Flash Attention in ~100 lines of CUDA (forward pass only)
Apache License 2.0
500 stars 40 forks source link