A reproducer to show that cuBLASLt appears to apply bias epilogue incorrectly when multiplying a float32 vector by a float32 matrix and the output matrix has ld > rows - View it on GitHub