Our work on Hardware-Efficient Attention for Fast Decoding has been accepted to COLM 2025. See you in Montreal !