Search papers, labs, and topics across Lattice.
Affiliation:
1
0
2
Training LLMs with tokens selected by gradient magnitude rather than just entropy boosts reasoning performance across diverse tasks.
