STAPO: Stabilizing Reinforcement Learning for LLMs by Silencing Rare Spurious Tokens

• URL: http://arxiv.org/abs/2602.15620v3
• Date: Mon, 23 Feb 2026 06:22:49 GMT
• Title: STAPO: Stabilizing Reinforcement Learning for LLMs by Silencing Rare Spurious Tokens
• Abstract summary: ExistingReinforcement Learning (RL) fine-tuning methods rely heavily on entropy regularization and reweighting to maintain stability.<n>In practice, they often suffer from late-stage performance collapse, leading to degraded reasoning quality and unstable training.<n>We find that training instability can be caused by a tiny fraction of tokens, approximately 0.01%, which we term spurious tokens.<n>We propose Spurious-Token-Aware Policy Optimization (STAPO), which promotes stable and effective large-scale model refinement.
• Score: 38.425692691443764
• License: http://creativecommons.org/licenses/by-nc-nd/4.0/
• Abstract: Reinforcement Learning (RL) has significantly improved large language model reasoning, but existing RL fine-tuning methods rely heavily on heuristic techniques such as entropy regularization and reweighting to maintain stability. In practice, they often suffer from late-stage performance collapse, leading to degraded reasoning quality and unstable training. Our analysis shows that the magnitude of token-wise policy gradients in RL is negatively correlated with token probability and local policy entropy. We find that training instability can be caused by a tiny fraction of tokens, approximately 0.01%, which we term spurious tokens. When such tokens appear in correct responses, they contribute little to the reasoning outcome but inherit the full sequence-level reward, leading to abnormally amplified gradient updates. To mitigate this instability, we design an S2T (silencing spurious tokens) mechanism to efficiently identify spurious tokens through characteristic signals with low probability, low entropy, and positive advantage, and then suppress their gradient perturbations during optimization. Incorporating this mechanism into a group-based objective, we propose Spurious-Token-Aware Policy Optimization (STAPO), which promotes stable and effective large-scale model refinement. Across six mathematical reasoning benchmarks using Qwen 1.7B, 8B, and 14B base models, STAPO consistently demonstrates superior entropy stability and achieves an average performance improvement of 7.13% ($ρ_{\mathrm{T}}$=1.0, top-p=1.0) and 3.69% ($ρ_{\mathrm{T}}$=0.7, top-p=0.9) over GRPO, 20-Entropy, and JustRL.

Related papers

• Which Tokens Matter? Adaptive Token Selection for RLVR with the Relative Surprisal Index [6.642733744927882]
  We argue that evaluating sampled-token probability or entropy in isolation is insufficient to capture the policy optimization dynamics.<n>We introduce the Relative Surprisal Index (RSI), a principled, information-theoretic metric that naturally couples the token's entropy with the probability of the selected token.<n>We show that RSI is related to the local ratio between the first-order variations of the logit-gradient norm and predictive entropy under a selected-logit perturbation.
  arXiv  Detail & Related papers  (2026-06-30T12:33:19Z)
• Beyond Entropy: Learning from Token-Level Distributional Deviations for LLM Reasoning [36.34371525731833]
  We introduce the Independent Combinatorial Tokens (ICT) framework, which shifts the optimization focus from scalar uncertainty to the distributional properties of token logits.<n>Our theoretical analysis, grounded in both Shannon and second-order Rényi entropy, proves that selectively updating on these tokens regulates policy concentration.<n> Empirical results demonstrate that updating only the top 10% of unique tokens on Qwen2.5 (0.5B/1.5B/7B) models yields an average pass@4 improvement of 4.58%, with a maximum gain of 14.9%, over GRPO, 20-Entropy, and STAPO baselines.
  arXiv  Detail & Related papers  (2026-06-18T04:11:56Z)
• STARE: Surprisal-Guided Token-Level Advantage Reweighting for Policy Entropy Stability [78.63876433996107]
  We conduct a first-order gradient analysis of token-level entropy dynamics under GRPO.<n>We propose STARE, which identifies entropy-critical token subsets via batch-internal surprisal quantiles.<n>On AIME24 and AIME25, STARE outperforms DAPO and other competitive baselines by 4%-8% in average accuracy.
  arXiv  Detail & Related papers  (2026-06-17T16:13:42Z)
• Not All Tokens Learn Alike: Attention Entropy Reveals Heterogeneous Signals in RL Reasoning [11.322734738973603]
  Reinforcement-learning-based post-training has become a key approach for improving the reasoning ability of large language models.<n>This work studies their heterogeneity through attention entropy, which measures how concentrated or diffuse the contextual support is for each response token.
  arXiv  Detail & Related papers  (2026-05-08T12:31:28Z)
• REAL: Regression-Aware Reinforcement Learning for LLM-as-a-Judge [83.2858110368572]
  We propose textbfREAL (underlineREgression-underlineAware Reinforcement underlineLThought), a principled RL framework designed to optimize regression rewards.<n>We show that REAL consistently outperforms both regression-aware SFT baselines and standard RL methods.
  arXiv  Detail & Related papers  (2026-03-17T21:19:08Z)
• From $\oldsymbol{\logπ}$ to $\oldsymbolπ$: Taming Divergence in Soft Clipping via Bilateral Decoupled Decay of Probability Gradient Weight [6.07556923512707]
  Reinforcement Learning with Verifiable Rewards (RLVR) has catalyzed a leap in Large Language Model (LLM) reasoning, yet its optimization dynamics remain fragile.<n>Standard algorithms like GRPO enforce stability via hard clipping'', which inadvertently stifles exploration by discarding gradients of tokens outside the trust region.<n>We propose Decoupled Gradient Policy Optimization (DGPO), which employs a decoupled decay mechanism based on importance sampling ratios.
  arXiv  Detail & Related papers  (2026-03-15T14:00:48Z)
• Online Causal Kalman Filtering for Stable and Effective Policy Optimization [23.37041897899078]
  We show that local off-policy deviation is structurally inconsistent at the token level.<n>We propose Online Causal Kalman Filtering for stable and effective Policy Optimization.
  arXiv  Detail & Related papers  (2026-02-11T07:57:43Z)
• Back to Basics: Revisiting Exploration in Reinforcement Learning for LLM Reasoning via Generative Probabilities [10.235183326885794]
  Reinforcement Learning with Verifiable Rewards (RLVR) has emerged as an indispensable paradigm for enhancing reasoning in Large Language Models (LLMs)<n>We analyze this issue from the perspective of sampling probability dynamics, identifying that the standard objective disproportionately reinforces the highest-likelihood paths.<n>We propose a novel Advantage Re-weighting Mechanism (ARM) designed to equilibrate the confidence levels across all correct responses.
  arXiv  Detail & Related papers  (2026-02-05T04:06:55Z)
• 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)
• A Step Back: Prefix Importance Ratio Stabilizes Policy Optimization [58.116300485427764]
  Reinforcement learning post-training can elicit reasoning behaviors in large language models.<n> token-level correction often leads to unstable training dynamics when the degree of off-policyness is large.<n>We propose a simple yet effective objective, Minimum Prefix Ratio (MinPRO)
  arXiv  Detail & Related papers  (2026-01-30T08:47:19Z)
• Principled RL for Diffusion LLMs Emerges from a Sequence-Level Perspective [85.06838178922791]
  Reinforcement Learning (RL) has proven highly effective for autoregressive language models.<n>But adapting these methods to diffusion large language models (dLLMs) presents fundamental challenges.<n>We propose a principled RL framework that treats entire sequence generation as a single action and uses the ELBO as a tractable sequence-level likelihood proxy.
  arXiv  Detail & Related papers  (2025-12-03T13:05:32Z)
• Revisiting Entropy in Reinforcement Learning for Large Reasoning Models [54.96908589622163]
  We investigate the entropy dynamics of large language models trained withReinforcement learning with verifiable rewards (RLVR)<n>Our findings reveal that the number of off-policy updates, the diversity of training data, and the clipping thresholds in the optimization objective are critical factors influencing the entropy of LLMs trained with RLVR.
  arXiv  Detail & Related papers  (2025-11-08T12:50:41Z)
• BAPO: Stabilizing Off-Policy Reinforcement Learning for LLMs via Balanced Policy Optimization with Adaptive Clipping [69.74252624161652]
  We propose BAlanced Policy Optimization with Adaptive Clipping (BAPO)<n>BAPO dynamically adjusts clipping bounds to adaptively re-balance positive and negative contributions, preserve entropy, and stabilize RL optimization.<n>On AIME 2024 and AIME 2025 benchmarks, our 7B BAPO model surpasses open-source counterparts such as SkyWork-OR1-7B.
  arXiv  Detail & Related papers  (2025-10-21T12:55:04Z)
• ASPO: Asymmetric Importance Sampling Policy Optimization [31.38346888572171]
  The Importance Sampling (IS) ratios of positive-advantage tokens are mismatched, leading to unbalanced token weighting for positive and negative tokens.<n>This mismatch suppresses the update of low-probability tokens while over-amplifying already high-probability ones.<n>We propose Asymmetric Importance Sampling Policy Optimization (ASPO), which uses a simple yet effective strategy that flips the IS ratios of positive-advantage tokens.
  arXiv  Detail & Related papers  (2025-10-07T15:54:24Z)
• From Uniform to Heterogeneous: Tailoring Policy Optimization to Every Token's Nature [38.46122853450324]
  Existing algorithms apply uniform optimization to all tokens, ignoring their different roles in reasoning process.<n>We introduce Heterogeneous Adaptive Policy Optimization (HAPO), a token-aware algorithm that dynamically adapts optimization based on token entropy.
  arXiv  Detail & Related papers  (2025-09-20T09:30:25Z)
• Beyond the 80/20 Rule: High-Entropy Minority Tokens Drive Effective Reinforcement Learning for LLM Reasoning [80.87085014818052]
  Reinforcement Learning with Verifiable Rewards (RLVR) has emerged as a powerful approach to enhancing the reasoning capabilities of Large Language Models (LLMs)<n>In this work, we undertake a pioneering exploration of RLVR through the novel perspective of token entropy patterns.<n>We observe that only a small fraction of tokens exhibit high entropy, and these tokens act as critical forks that steer the model toward diverse reasoning pathways.
  arXiv  Detail & Related papers  (2025-06-02T17:54:39Z)
• Do Not Let Low-Probability Tokens Over-Dominate in RL for LLMs [25.575582861331405]
  Low-probability tokens disproportionately influence model updates due to their large gradient magnitudes.<n>We propose two novel methods: Advantage Reweighting and Low-Probability Token Isolation (Lopti)<n>Our approaches promote balanced updates across tokens with varying probabilities, thereby enhancing the efficiency ofReinforcement learning.
  arXiv  Detail & Related papers  (2025-05-19T10:14:08Z)

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.