Search papers, labs, and topics across Lattice.
3
0
3
9
Learning physics into generative crystal models *after* training, rather than during inference, drastically improves structural validity and packing without sacrificing generation speed, and this benefit transfers across different model architectures.
VLSR's localize-then-reason approach boosts throughput by 9.6X, revolutionizing how we analyze molecular properties from images.
Mechanism-aware training boosts LLM performance in chemical reasoning, achieving an 8.3% exact match on complex reaction pathways鈥攐utperforming specialized models.
