Search papers, labs, and topics across Lattice.
National Yang Ming Chiao Tung University
1
0
2
LATCH accelerates diffusion language model decoding by up to 17.8x without compromising accuracy, redefining how we approach early exits in generative tasks.
