Dr. OPD: Learning What to Follow for Optimal On-Policy Distillation of Large Language Models B7CAM06Q77
| URI | m2://res/B7CAM06Q77 |
|---|---|
| Kind | preprint |
| Creators | Zhenyu Wang, Tianze Wang, Linjun Zhang, Yifan Hu |
| Date | 2026-09-29 |
| Subjects | arxiv cs.LG, arxiv cs.AI, arxiv cs.CL, arxiv math.OC |
| Scope | border |
Expressions
e1: version, preprint
| Date | 2026-09-29 |
|---|---|
| Ids | arXiv 2609.38025 |
Items in this resource
The mentions of the inventory that name this resource. Each name links to the card of its inventory row.
| Item | Kind | Where | Source |
|---|---|---|---|
| Reward improvement over vanilla OPD | statement | theorem 5 | arxiv |
| Weighted KL divergence | definition | definition 1 | arxiv |
| Validity of weighted OPD | statement | theorem 3 | arxiv |
| Weighted OPD | definition | definition 2 | arxiv |
| Lower bound on reward | statement | proposition 4 | arxiv |
Catalogue records
- arxiv: 2609.38025 (work)