Efficient Sparse Attention needs Adaptive Token Release.

Chaoran Zhang, Lixin Zou, Dan Luo, Xiangyang Luo, Zihao Li, Min Tang, Chenliang Li

Browse the full ACL paper archive.