Search papers, labs, and topics across Lattice.
Affiliation:
1
0
3
A learned "Token-to-Go" progress meter enables multi-turn RL agents to effectively learn from sparse rewards, boosting performance by up to 7.2 points without increasing training costs.
