Search papers, labs, and topics across Lattice.
1
0
2
By strategically amplifying updates along flat directions in the loss landscape, LITE unlocks faster LLM pre-training with existing matrix-based optimizers like Muon and SOAP.
