Inference & Serving · 2025
On-Policy Distillation
Combines on-policy sampling from the student with dense per-token teacher supervision, reaching RL-level reasoning-benchmark performance at a fraction of RL's compute cost.
Editorial record
Plain-language summary
Combines on-policy sampling from the student with dense per-token teacher supervision, reaching RL-level reasoning-benchmark performance at a fraction of RL's compute cost.
Knowledge graph
Relationships
Descendants
Depends onEvidence: Strongly supported
Distilling the Knowledge in a Neural Network
Combines on-policy sampling with knowledge distillation
TML
Source record
Provenance
- Record ID
- P-658
- Record created
- 2026-07-13
- Last reviewed
- 2026-07-14
- Record version
- 2
Citation caveat: Citation metadata is approximate and marked unverified in the source dataset.