paper / chhillee / May 14 / failedFlex Attention: A Programming Model for Generating Optimized Attention Kernels