Search papers, labs, and topics across Lattice.
1
0
2
Spectral preprocessing of query-key projections can reduce attention computation costs by up to 79% while preserving performance in character-level language modeling.
