Search papers, labs, and topics across Lattice.
University of British Columbia
1
0
2
d-OPSD enables dLLMs to learn from their own future outputs, drastically improving sample efficiency and performance in reasoning tasks.