Search papers, labs, and topics across Lattice.
1
0
3
2
Reduce your MoE LLM's memory bottleneck by up to 49% with a simple router fine-tuning trick that encourages experts to be reused.
