Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
Seoul National University
1
0
3
Transformer attention heads are shockingly low-rank: language models concentrate their attention into just a handful of dimensions, wasting most of the allocated capacity.