“OpenAI has implemented Flash Attention in Triton, a Python-like language for writing GPU kernels.”