Search papers, labs, and topics across Lattice.
1
0
3
0
Enterprise LLMs can achieve state-of-the-art performance with significantly fewer active parameters and tokens by using a Mixture-of-Experts architecture and a novel RL training method to reduce overthinking.
