Search papers, labs, and topics across Lattice.
Affiliation:
1
0
3
Distilling explicit, reward-guided skills into VLMs without the student model ever seeing the skills themselves unlocks superior multimodal performance.