Search papers, labs, and topics across Lattice.
Harbin Institute of Technology (ShenZhen)
3
0
5
6
Discarding just a few high-advantage rollouts during GRPO updates dramatically boosts LLM reasoning and exploration.
CRISP reduces interaction turns by up to 33% while maintaining accuracy by prioritizing critical evidence-gathering steps in deep search agents.
Distilled RL achieves a breakthrough in LLM post-training by effectively transferring previously inaccessible knowledge, outperforming traditional methods in both accuracy and adaptability.
