Search papers, labs, and topics across Lattice.
University of California, Los Angeles (UCLA)
1
0
2
Shifting the focus from token likelihood to target distribution design reveals a more effective framework for supervised fine-tuning that consistently outperforms traditional methods.
