Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
1
0
3
0
Forget LoRA: tuning just a single initial state matrix in hybrid recurrent-attention models delivers comparable or better performance with zero inference overhead.