Search papers, labs, and topics across Lattice.
1
0
3
Achieve Transformer-level handwritten text recognition accuracy with 1.6-1.9x faster inference and 38-42% less memory by ditching attention for a linear-time RetNet architecture.
