Search papers, labs, and topics across Lattice.
1
0
3
Stop LLMs from unlearning: SLIME stabilizes likelihood during preference optimization, preventing degradation of high-quality outputs and formatting collapse.
