Search papers, labs, and topics across Lattice.
3
0
4
5
Single-rollout sampling can dramatically improve the stability and effectiveness of RL training for large language models, outperforming traditional methods by a significant margin.
CompactionRL enables LLMs to effectively manage long-horizon tasks by summarizing context, leading to substantial performance gains in coding benchmarks.
GLM-5 doesn't just code; it engineers, showcasing unprecedented capability in tackling end-to-end software engineering challenges.
