Search papers, labs, and topics across Lattice.
4
1
8
3
Proactively committing mid-entropy pivot positions can accelerate dLLM decoding by up to 18 times while improving accuracy.
Training speculative decoding models just got an order of magnitude faster, unlocking real-world deployment with a new open-source framework and a suite of production-ready draft models.
Squeeze your image datasets without sacrificing model accuracy: DCQ intelligently reduces color redundancy, outperforming naive compression methods.
DPO's reliance on a reference policy can backfire, prematurely halting learning when the reference is pessimistically wrong, but a simple one-line fix can significantly improve performance.
