Search papers, labs, and topics across Lattice.
1
0
3
6
On-policy LLM post-training's generalization advantage stems from its dynamic parameter update direction, a principle transferable to SFT for improved performance and efficiency.
