Search papers, labs, and topics across Lattice.
3
0
7
AdaThinkV achieves 40.79% accuracy in video reasoning while using 22.7% fewer tokens than its strongest adaptive baseline, showcasing a breakthrough in token-efficient reasoning.
Don't let shallow-layer cues fool you: SAFE-Pruner uses "future sight" to prune vision-language-action models, achieving 1.89x speedup without sacrificing task success.
LLMs can substantially improve their ability to follow complex instructions and constraints by explicitly auditing their own context adherence before answering.
