5arXiv cs.CL (Computation and Language)·
BIRD: Bootstrapped Iterative Self-Reasoning Distillation reduces LLM chain-of-thought length while improving accuracy
Researchers introduce BIRD (Bootstrapped Iterative Self-Reasoning Distillation), a two-stage method for compressing chain-of-thought reasoning in large language models without sacrificing accuracy. The approach first fine-tunes a model on brevity-instructed correct traces to warm-start the rollout distribution, then applies on-policy reverse-KL distillation against a concise self-teacher. On Qwen3-8B, BIRD improves MATH-500 accuracy from 86.2% to 92.0% while cutting average response length from 3,099 to 1,115 tokens, outperforming cold-start on-policy distillation baselines.
Related guides (3)
Related events (8)
Purified OPSD fixes on-policy self-distillation failures in long chain-of-thought reasoning models
A new arXiv preprint identifies why on-policy self-distillation (OPSD) consistently degrades long chain-of-thought reasoning models: the teacher's supervision signal is dominated by reference-induced shortcuts rather than question-conditioned, transferable corrections. The authors propose a two-step fix using a reference-only teacher to isolate the non-transferable component and pointwise mutual information (PMI) to construct a cleaner distillation target. Experiments across four long-CoT models on two datasets show consistent improvements over both the base model and standard OPSD while preserving reflective reasoning behavior.
Step-aligned critique outperforms GRPO and reference-solution conditioning in self-distillation
A new arXiv paper investigates context design for self-distillation of language models, comparing binary reward (GRPO), reference solutions, and step-by-step critiques aligned to the solver's reasoning trace. Step-aligned critique yields the largest gains, outperforming GRPO by 16.11 points and reference-solution conditioning by 5.27 points on Avg@12. Per-token advantage analysis shows that step-aligned feedback targets only failing tokens, avoiding unnecessary pressure on already-correct reasoning steps. The findings suggest structural alignment between feedback and the model's reasoning trace is a key driver of self-distillation effectiveness.
On-Policy Delta Distillation improves reasoning transfer via teacher-base model difference signal
Researchers from NAVER AI introduce On-Policy Delta Distillation (OPD²), a new post-training method that replaces direct imitation of a teacher model's output distribution with a 'delta signal' — the difference between the teacher and its pre-instruction-tuning base model. This delta signal isolates the reasoning capability changes induced by instruction tuning, providing a more targeted supervision signal for student models. Experiments across math, science, and code-reasoning benchmarks show OPD² consistently outperforms conventional on-policy distillation with shorter post-training periods.
Skill-Conditioned Gated Self-Distillation (SGSD) for LLM Reasoning
SGSD is a new on-policy self-distillation method for LLM reasoning that replaces trusted privileged information (e.g., reference answers) with an experience-derived skill bank of skill-mistake pairs. It constructs a multi-teacher pool, validates each teacher's contribution via a verifier, and applies a gated objective to distill informative disagreements while suppressing noisy signals. On Qwen3-1.7B, SGSD outperforms GRPO by 6.2% and answer-conditioned OPSD by 1.7% on average across AIME24, AIME25, and HMMT25. The method relaxes the assumption of trusted privileged information, making self-distillation more practical under weaker supervision.
Rubric-Conditioned Self-Distillation: structured feedback for reasoning model post-training
A new arXiv preprint proposes Rubric-Conditioned Self-Distillation (RCSD), a post-training framework that replaces scalar reward signals and noisy chain-of-thought annotations with structured rubrics for fine-grained credit assignment. The method conditions a teacher model on criterion-level rubrics to provide token-level guidance on the student's own sampled trajectories, avoiding reliance on a single reference rationale. Evaluated on science reasoning benchmarks, RCSD outperforms GRPO by 1.0 points and OPSD by 0.9 points on average.
Are Full Rollouts Necessary for On-Policy Distillation?
This paper investigates whether full rollouts are required during on-policy distillation (OPD) for training reasoning models, identifying rollout horizon as a key computational bottleneck. The authors propose two strategies: Progressive OPD (POPD), which gradually expands rollout horizon during training, and Truncated OPD (TOPD), which uses permanently truncated rollouts. Experiments on mathematical reasoning show POPD achieves up to 3× training efficiency improvement, while TOPD matches full OPD performance using only 10% of the rollout horizon, yielding significant wall-clock and memory savings.
MADA-RL: Multi-agent debate with counterfactual critic advantage for parameter-efficient reasoning in compact models
MADA-RL is a post-training framework that specializes compact LLMs (≤4B parameters) into generator and critic roles, training them with a debate-aware RL signal using LoRA adapters on only a small fraction of parameters. The central contribution is a counterfactual critic advantage that conditions the critic's reward on the generator ensemble's per-instance accuracy, explicitly incentivizing critics to correct generator errors rather than reproduce correct answers. Applied to DeepSeek-R1-Distill-Qwen-1.5B, the method achieves +2.0 percentage points on mathematical reasoning benchmarks using 16× fewer trainable parameters than fully fine-tuned baselines, placing it on the accuracy-vs-trainable-parameter Pareto front. The approach does not surpass the strongest baselines (DeepScaleR, STILL-3) trained on larger datasets, and the paper analyzes this gap directly.
Strong Teacher Not Needed? On Distillation in LLM Pretraining
This paper challenges the conventional assumption that knowledge distillation requires a stronger teacher to produce better students. Through systematic variation of architecture sizes and training token budgets, the authors find that even small, undertrained teachers can improve larger student models when language modeling and distillation losses are properly mixed. Counterintuitively, stronger teachers can saturate or reverse distillation gains, and distillation benefits generalization more than in-domain fitting.