# Trapped by Their Own Rollouts: Understanding Aggregation--Rollout Feedback in Federated On-Policy Distillation

arXiv: https://arxiv.org/abs/2609.32573 · HTML: https://mlsanity.com/paper/2609.32573/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in On-policy and self-distillation for LLMs, one of 12 directions in Knowledge distillation.
- Builds on 19 of its 23 references that are in the map.
- First posted on arXiv 26 Sep 2026.

## Placed in

- Knowledge distillation / [On-policy and self-distillation for LLMs](https://mlsanity.com/distill/on-policy-and-self-distillation-for-llms/)

## Builds on (most cited first, 19 of 19)

- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [Communication-Efficient Learning of Deep Networks from Decentralized Data](https://mlsanity.com/paper/1602.05629/) (2016)
- [Measuring Mathematical Problem Solving With the MATH Dataset](https://mlsanity.com/paper/2103.03874/) (2021)
- [Trust Region Policy Optimization](https://mlsanity.com/paper/1502.05477/) (2015)
- [Solving Quantitative Reasoning Problems with Language Models](https://mlsanity.com/paper/2206.14858/) (2022)
- [OlympiadBench: A Challenging Benchmark for Promoting AGI with Olympiad-Level Bilingual Multimodal Scientific Problems](https://mlsanity.com/paper/2402.14008/) (2024)
- [On-Policy Distillation of Language Models: Learning from Self-Generated Mistakes](https://mlsanity.com/paper/2306.13649/) (2023)
- [Performative Prediction](https://mlsanity.com/paper/2002.06673/) (2020)
- [Rethinking On-Policy Distillation of Large Language Models: Phenomenology, Mechanism, and Recipe](https://mlsanity.com/paper/2604.13016/) (2026)
- [Entropy-Aware On-Policy Distillation of Language Models](https://mlsanity.com/paper/2603.07079/) (2026)
- [Improving LoRA in Privacy-preserving Federated Learning](https://mlsanity.com/paper/2403.12313/) (2024)
- [FLoRA: Federated Fine-Tuning Large Language Models with Heterogeneous Low-Rank Adaptations](https://mlsanity.com/paper/2409.05976/) (2024)
- [FedExP: Speeding Up Federated Averaging via Extrapolation](https://mlsanity.com/paper/2301.09604/) (2023)
- [Finite-Time Analysis of On-Policy Heterogeneous Federated Reinforcement Learning](https://mlsanity.com/paper/2401.15273/) (2024)
- [Rethinking On-Policy Distillation of Large Language Models II: One Training Example](https://mlsanity.com/paper/2609.04172/) (2026)
- [Learning to Foresee: Unveiling the Unlocking Efficiency of On-Policy Distillation](https://mlsanity.com/paper/2605.11739/) (2026)
- [Performative Federated Learning: A Solution to Model-Dependent and Heterogeneous Distribution Shifts](https://mlsanity.com/paper/2305.05090/) (2023)
- [What Matters in On-Policy Distillation? A Perspective on Data Efficiency and Data Selection](https://mlsanity.com/paper/2609.05198/) (2026)
- [Convergence Guarantees for Federated SARSA with Local Training and Heterogeneous Agents](https://mlsanity.com/paper/2512.17688/) (2025)

## Abstract (arXiv)

On-policy distillation (OPD) is a promising approach to language-model adaptation, aligning teacher supervision with the student's own generated trajectories. When adaptation prompts are distributed across clients, can this process benefit from federated collaboration? We study federated OPD and find that substantial collaboration gains can be obscured by learning-rate sensitivity: FedAvg can perform no better than independent local training at a small learning rate, yet recover a clear advantage at a larger rate. We explain this phenomenon through the student's dual role as learner and generator of future training data. An aggregation-induced optimization lag can delay access to useful teacher supervision, which in turn slows subsequent learning. Our theory establishes this aggregation--rollout feedback in a solvable model with a common optimum and stable updates, and identifies two coupled roles of learning rate: learning from current supervision and reaching future supervision. Guided by this analysis, we propose FedTOPS (Federated Teacher-guided On-Policy Scaling), which reuses teacher feedback on current trajectories to adapt the FedAvg update magnitude under clientwise predictive-change constraints. Across six mathematical reasoning benchmarks, FedTOPS improves macro Avg@8 over FedAvg by 4.56--14.57 percentage points across the evaluated student models and local learning rates.