If so, when are we going to have it? I have to ask the team because flash-attention is not a model implementation. It is the implementation of matrix multiplication.
