Restoring the Sweet Spot: Pass-Rate Weighted Self-Distillation for LLM Reasoning

• URL: http://arxiv.org/abs/2605.27765v1
• Date: Tue, 26 May 2026 23:30:01 GMT
• Title: Restoring the Sweet Spot: Pass-Rate Weighted Self-Distillation for LLM Reasoning
• Abstract summary: Self-Distillation Policy Optimization (SDPO) provides dense token-level credit assignment for reinforcement learning with large language models.<n>We analyze this gap through the lens of GRPO's advantage normalization.<n> Experiments on scientific reasoning and tool-use benchmarks demonstrate that SC-SDPO consistently improves over SDPO.
• Score: 37.935234078277354
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Self-Distillation Policy Optimization (SDPO) provides dense token-level credit assignment for reinforcement learning with large language models by leveraging the model's own feedback-conditioned predictions as a self-teacher. Unlike GRPO, however, whose group-relative advantage naturally concentrates learning on a sweet spot of intermediate-difficulty questions, SDPO's KL-based advantage lacks an implicit notion of difficulty awareness. We analyze this gap through the lens of GRPO's advantage normalization. Extending the learnability framework to normalized rewards, we show that normalization absorbs the variance term $p(1-p)$, equalizing leading-order learnability across questions and leaving $\sqrt{p(1-p)}$ as the sole residual scaling factor in the per-question gradient. This analysis yields a simple prescription: weight each question's SDPO loss by $[\hat{p}(1-\hat{p})]^{1/2}$, resulting in SC-SDPO, a scale-consistent variant of SDPO. The proposed weights are obtained as a zero-cost byproduct of on-policy rollouts with batch-adaptive normalization, inducing an implicit curriculum that dynamically tracks the model's evolving competence. Experiments on scientific reasoning and tool-use benchmarks demonstrate that SC-SDPO consistently improves over SDPO, yielding gains of +3.2/+4.3 (mean@16/maj@16) on Qwen3-8B and +1.8/+3.0 on OLMo-3-7B, while preserving stable training dynamics throughout optimization.

Related papers

• Dropout-GRPO: Variational Stochasticity for Continuous Latent Reasoning [0.0]
  latent-reasoning models like Coconut feed continuous hidden states recurrently in place of discrete chain-of-thought tokens.<n>Applying group-relative reinforcement learning to continuous latent reasoning has proven difficult.<n>By applying a single Bernoulli mask held constant across all latent recurrence steps for a given rollout, we generate essential trajectory variance.
  arXiv  Detail & Related papers  (2026-06-08T21:21:42Z)
• The Easy, the Hard, and the Learnable: Confidence and Difficulty-Adaptive Policy Optimization for LLM Reasoning [74.8743478567224]
  GRPO-style training treats easy, hard, and learnable questions alike through uniform sampling and weighting.<n>We propose Confidence and Difficulty-adaptive Policy Optimization (CoDaPO), which assigns each question a bounded value from rollout confidence and empirical difficulty.<n>Across twelve benchmarks, CoDaPO consistently improves accuracy over existing RL methods.
  arXiv  Detail & Related papers  (2026-06-06T02:51:33Z)
• Physics-Guided Policy Optimization with Self-Distillation [13.212487988108592]
  Self-distilled policy optimization (SDPO) has become a popular paradigm for LLM post-training.<n>We propose Physics-Guided Policy Optimization (PGPO), which introduces an information-modulated step-size multiplier.<n>We show that this modulation preserves the order-1 weak-approximation guarantees of vanilla SGD, and incurs negligible overhead per iteration.
  arXiv  Detail & Related papers  (2026-06-02T13:20:39Z)
• ODRPO: Ordinal Decompositions of Discrete Rewards for Robust Policy Optimization [14.900223489465683]
  Large Language Models (LLMs) utilize Reinforcement Learning from AI Feedback (RLAIF) for non-verifiable domains.<n>These domains often rely on based auto-raters to provide granular, multi-tier discrete rewards.<n>We introduce $textbfO$ptimization ($textbfOOO$), a framework that structurally isolates evaluation noise by decomposing discrete rewards into a sequence of ordinal binary indicators.
  arXiv  Detail & Related papers  (2026-05-12T19:17:14Z)
• Your Language Model is Its Own Critic: Reinforcement Learning with Value Estimation from Actor's Internal States [14.658333957936769]
  We introduce Policy Optimization with Internal State Value Estimation.<n>It obtains a baseline at negligible cost by using the policy model's internal signals already computed during the policy forward pass.<n>A lightweight probe predicts the expected verifiable reward from the hidden states of the prompt and generated trajectory.
  arXiv  Detail & Related papers  (2026-05-08T10:49:36Z)
• Rethinking Importance Sampling in LLM Policy Optimization: A Cumulative Token Perspective [22.848847562976633]
  Existing methods face a fundamental bias-variance dilemma: token-level IS ratios, as adopted by PPO and GRPO, introduce bias by ignoring prefix state distribution mismatch.<n>We propose CTPO, which combines the cumulative token IS ratio with position-adaptive clipping that scales log-space clip bounds according to the natural $sqrtt$ growth of the cumulative log-ratio.
  arXiv  Detail & Related papers  (2026-05-08T06:35:02Z)
• Too Correct to Learn: Reinforcement Learning on Saturated Reasoning Data [55.84428098924793]
  We propose Constrained Uniform Top-K Sampling (CUTS), a parameter-free decoding strategy enforcing structure-preserving exploration.<n>We integrate this into Mixed-CUTS, a training framework synergizing exploitative and exploratory rollouts to amplify intra-group advantage variance.<n> Notably, Mixed-CUTS improves Pass@1 accuracy on the challenging AIME25 benchmark by up to 15.1% over standard GRPO.
  arXiv  Detail & Related papers  (2026-04-20T16:43:28Z)
• Unifying Group-Relative and Self-Distillation Policy Optimization via Sample Routing [79.88256756334327]
  Self-Distillation Policy Optimization (SDPO) addresses this by providing denser, more targeted logit-level supervision.<n>We propose Sample-Routed Policy Optimization (SRPO)<n>SRPO routes correct samples to GRPO's reward-aligned reinforcement and failed samples to SDPO's targeted logit-level correction.
  arXiv  Detail & Related papers  (2026-04-02T17:29:18Z)
• Improving Search Agent with One Line of Code [68.58667107354253]
  Tool-based Agentic Reinforcement Learning (TARL) has emerged as a promising paradigm for training search agents to interact with external tools.<n>We propose textbfSearch textbfAgent textbfPolicy textbfOptimization (textbfSAPO), which stabilizes training via a conditional token-level KL constraint.
  arXiv  Detail & Related papers  (2026-03-10T04:07:39Z)
• Unbiased Dynamic Pruning for Efficient Group-Based Policy Optimization [60.87651283510059]
  Group Relative Policy Optimization (GRPO) effectively scales LLM reasoning but incurs prohibitive computational costs.<n>We propose Dynamic Pruning Policy Optimization (DPPO), a framework that enables dynamic pruning while preserving unbiased gradient estimation.<n>To mitigate the data sparsity induced by pruning, we introduce Dense Prompt Packing, a window-based greedy strategy.
  arXiv  Detail & Related papers  (2026-03-04T14:48:53Z)
• Rethinking the Trust Region in LLM Reinforcement Learning [72.25890308541334]
  Proximal Policy Optimization (PPO) serves as the de facto standard algorithm for Large Language Models (LLMs)<n>We propose Divergence Proximal Policy Optimization (DPPO), which substitutes clipping with a more principled constraint.<n>DPPO achieves superior training and efficiency compared to existing methods, offering a more robust foundation for RL-based fine-tuning.
  arXiv  Detail & Related papers  (2026-02-04T18:59:04Z)
• Single-stream Policy Optimization [21.214853668053234]
  We introduce Single-stream Policy Optimization (SPO)<n>SPO replaces per-group baselines with a persistent, KL-adaptive value tracker and normalizes advantages globally across the batch.<n>Being group-free, SPO enables higher throughput and scales effectively in long-horizon or tool-integrated settings.
  arXiv  Detail & Related papers  (2025-09-16T16:39:11Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.