Search papers, labs, and topics across Lattice.
3
0
5
5
Single-rollout sampling can dramatically enhance the stability and effectiveness of RL in large language models, outperforming traditional methods in challenging benchmarks.
CompactionRL enables LLMs to effectively manage long-horizon tasks by summarizing context, leading to substantial performance gains in coding benchmarks.
Multimodal perception is no longer just an add-on: GLM-5V-Turbo bakes it directly into the core of reasoning, planning, and action.
