arXiv Machine Learning
Progress-Aware Distillation for Mitigating Stagnation in Small Language Model Agents
Read the original on arXiv Machine Learning →The Flow has not summarised this story yet — read it at arXiv Machine Learning.
Related stories
arXiv AI
Aug 19
SOD: Step-wise On-policy Distillation for Small Language Model Agents
SOD: Step-wise On-policy Distillation for Small Language Model Agents proposes a new framework that adaptively reweights distillation strength at each reasoning step based on step-level divergence. This approach mitigates cascading errors in tool-integrated reasoning by attenuating misleading teacher signals in high-divergence regions while preserving dense guidance where student and teacher align. Experiments on math, science, and code benchmarks show up to 20.86% improvement over the second-best baseline, with a 0.6B student scoring 26.13% on AIME 2025.
arXiv AI
Jul 24
Student-Centered Distillation Narrows the Agentic Gap Between Small and Large LLMs
arXiv:2509. 14257v3 Announce Type: replace-cross Abstract: Large Language Model agents achieve strong performance on multi-step reasoning and tool-use tasks, but their impressive capabilities typically rely on extremely large backbones.
arXiv AI
Oct 5
Harness-Aware Distillation for Small Language Model Agents
The paper introduces Harness-Aware Distillation (HAD), a method for training smaller language model agents that preserves the surrounding harness—software managing context, tools, and feedback—while focusing distillation on the teacher’s contributions beyond the harness. HAD combines an action preference that contrasts teacher actions with and without harness information, and a validity check that filters out contradictory preference pairs. Experiments on long-horizon agent benchmarks show that HAD outperforms standard on‑policy distillation, reducing unproductive loops and improving error recovery without requiring task rewards or future information.
arXiv AI
Jun 19
Reinforcement-aware Knowledge Distillation for LLM Reasoning
arXiv:2602. 22495v3 Announce Type: replace-cross Abstract: Reinforcement learning (RL) post-training has recently driven major gains in long chain-of-thought reasoning large language models (LLMs), but the high inference cost of such models motivates distillation into smaller students.
arXiv Machine Learning
Oct 8
TAP: Efficient Long-Horizon Agent Pruning via Trajectory-Anchored Recovery
TAP (Trajectory‑Anchored Pruning) is a structural pruning framework designed for reinforcement‑learning‑trained agents that addresses the shortcomings of existing pruning methods on long‑horizon tasks. It couples iterative channel pruning with on‑policy recovery, anchoring student interactions to teacher trajectories and re‑scoring channels using gradients from the recovery objective. Experiments show that removing 60% of FFN channels preserves 99.2% and 88.0% of task success rates on ALFWorld and WebShop while cutting GPU time per successful task by about 22% and 17%.
arXiv AI
1d ago
D-Skill: Distilling Teacher Knowledge through Agent Skills
D‑Skill is a framework that distills the behavioral advantage of a large language‑model teacher into an interpretable natural‑language skill document, using only paired rollouts and no weight updates. The skill is constructed offline from fixed trajectories, then provided to a smaller student model at inference time. Across benchmarks such as SpreadsheetBench‑Verified, ALFWorld, and chess puzzles, D‑Skill closes 75‑95% of the teacher‑student performance gap and reduces inference cost by 5.7×–8.3× compared to the teacher.