- Rating
- 2
- Confidence
- 2
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
This paper proposes Capability-Aware Multi-Teacher On-Policy Distillation (CA-MOPD). It first initializes a student by applying Task Arithmetic to domain-specific RL specialists, then adjusts domain-level distillation weights using verifier-estimated teacher–student capability gaps and a regression guard based on the student’s historical-best capability. The underlying dense token-level distillation objective remains unchanged. Experiments use a SmolLM3-3B model across mathematics, code generation, and instruction following. CA-MOPD outperforms naive MOPD, Open-MOPD, and the merged initialization in six-benchmark average.
Critical strengths
- Clear motivation. The distinction between behavioral discrepancy and task-level capability advantage is relevant, particularly when specialist merging has already transferred substantial capability.
- Useful methodological detail. The appendices describe the controller, optimization objective, training configuration, and state required for checkpoint resumption.
Critical weaknesses
- Incomplete contribution isolation. The main experiments do not sufficiently separate the benefits of initialization, capability-based allocation, regression retention, token balancing, and global loss scaling.
- Insufficient validation of the capability signal. The controller compares a fixed teacher reference with student estimates from small, changing training batches, without demonstrating that these estimates reliably distinguish capability changes from sampling variation.
- Limited evidence of robustness. The reported margins lack uncertainty estimates or independent training-seed results.
Questions for authors
Based on the review above and some other issues, the reviewer has the following questions or comments.
- Were the Table 1 baselines (especially the MOPD ones) initialized from the same merged checkpoint?
- Could the authors provide the actual ablation experimental results of different CA-MOPD variants, including CA-MOPD without merged initialization, CA-MOPD without capability-based allocation, CA-MOPD without regression retention, CA-MOPD without token balancing, CA-MOPD without global loss scaling? These can help justify the efficacy of each individual component, especially those not already equipped in Naive MOPD and Open-MOPD.
- Equation (3) defines capability under a common domain distribution. The implemented estimator instead compares a teacher score fixed on a held-out calibration set with an EMA of student scores on changing training batches. How comparable are the fixed teacher calibration scores and student estimates from changing training batches? Please specify the calibration-data source and selection procedure, clarify whether calibration decoding matches online training decoding, and report the initial teacher/student calibration scores with uncertainty. A fixed, matched teacher–student probe at selected checkpoints would help validate the online signal.
- Are the gains reproducible across training seeds? Please report uncertainty for the main comparisons; repeated decoding alone does not measure training variability.