Search papers, labs, and topics across Lattice.
DualityRL
1
0
3
Pruning reasoning paths with a learned "STOP" token slashes compute costs and boosts accuracy in large reasoning models, outperforming existing methods.
