Search papers, labs, and topics across Lattice.
1
0
3
0
By prioritizing diversity over accuracy in experience replay, DyJR significantly boosts LLM reasoning performance in RL, outperforming GRPO and other baselines without sacrificing training efficiency.
