Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
3
0
3
0
Trillion-token scale looped MoE models achieve equivalent reasoning performance to 2x larger non-looped MoE models with matched training compute, unlocking test-time scaling via recurrence.
On-device LLMs can achieve state-of-the-art performance with significantly reduced computational cost by leveraging a carefully designed Mixture-of-Experts architecture, challenging the assumption that dense models are always superior for mobile deployment.