Search papers, labs, and topics across Lattice.
Kyoto University
1
0
3
Forget expensive distillation – aligning language models can be as simple as carefully choosing the right mix of pretraining data based on log-likelihood differences.
