Dissecting Long-Chain-of-Thought Reasoning Models: An Empirical Study

• URL: http://arxiv.org/abs/2506.04913v2
• Date: Mon, 10 Nov 2025 12:38:29 GMT
• Title: Dissecting Long-Chain-of-Thought Reasoning Models: An Empirical Study
• Authors: Yongyu Mu, Jiali Zeng, Bei Li, Xinyan Guan, Fandong Meng, Jie Zhou, Tong Xiao, Jingbo Zhu,
• Abstract summary: This work focuses on the roles of positive and negative samples in scaling reinforcement learning.<n>We identify substantial data inefficiency in group relative policy optimization, where over half of the samples yield zero advantage.<n>We investigate unstable performance across various reasoning models and benchmarks, attributing instability to uncertain problems with ambiguous outcomes.
• Score: 91.78803511141975
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Despite recent progress in training long-chain-of-thought reasoning models via scaling reinforcement learning (RL), its underlying training dynamics remain poorly understood, and several counterintuitive behaviors persist. This work focuses on three key aspects: (1) We systematically analyze the roles of positive and negative samples in scaling RL, revealing that positive samples mainly facilitate precise fitting to the training data, whereas negative samples significantly enhance generalization and robustness. Interestingly, while positive samples are essential for convergence in the zero-RL setting, training on negative samples alone suffices to attain strong reasoning performance and even better generalization in cold-start scenarios. (2) We identify substantial data inefficiency in group relative policy optimization, where over half of the samples yield zero advantage. To address this, we explore two strategies, including relative length rewards and offline sample injection, to leverage these data better and enhance reasoning efficiency and capability. (3) We investigate unstable performance across various reasoning models and benchmarks, attributing instability to uncertain problems with ambiguous outcomes, and demonstrate that greedy decoding can distort evaluation by flipping the correctness of responses. Our code is available at: https://github.com/takagi97/Dissect-Long-Reason-Models.

Related papers

• Not All Negative Samples Are Equal: LLMs Learn Better from Plausible Reasoning [23.352305094623123]
  Plausible Negative Samples (PNS) is a method that synthesizes high-quality negative samples exhibiting expected format and structural coherence.<n>PNS trains a dedicated model via reverse reinforcement learning guided by a composite reward combining format compliance, accuracy inversion, and chain-of-thought evaluation.<n>PNS consistently outperforms other negative sample methods, achieving an average improvement of 2.03% over RL-trained models.
  arXiv  Detail & Related papers  (2026-02-03T13:32:02Z)
• 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)
• Beyond Accuracy: Dissecting Mathematical Reasoning for LLMs Under Reinforcement Learning [93.00629872970364]
  Reinforcement learning (RL) has become the dominant paradigm for improving the performance of language models on complex reasoning tasks.<n>We introduce SPARKLE, a fine-grained analytic framework to dissect the effects of RL across three key dimensions.<n>We study whether difficult problems -- those yielding no RL signals and mixed-quality reasoning traces -- can still be effectively used for training.
  arXiv  Detail & Related papers  (2025-06-05T07:53:59Z)
• Model-agnostic Mitigation Strategies of Data Imbalance for Regression [0.0]
  Data imbalance persists as a pervasive challenge in regression tasks, introducing bias in model performance and undermining predictive reliability.<n>We present advanced mitigation techniques, which build upon and improve existing sampling methods.<n>We demonstrate that constructing an ensemble of models -- one trained with imbalance mitigation and another without -- can significantly reduce these negative effects.
  arXiv  Detail & Related papers  (2025-06-02T09:46:08Z)
• Behavior Injection: Preparing Language Models for Reinforcement Learning [24.46625106928253]
  Reinforcement fine-tuning (RFT) has emerged as a powerful post-training technique to incentivize the reasoning ability of large language models (LLMs)<n>LLMs can respond very inconsistently to RFT: some show substantial performance gains, while others plateau or even degrade.<n>We propose behavior injection, a task-agnostic data-augmentation scheme applied prior to RL.
  arXiv  Detail & Related papers  (2025-05-25T00:54:50Z)
• Unearthing Gems from Stones: Policy Optimization with Negative Sample Augmentation for LLM Reasoning [48.33401015101481]
  We propose Behavior Constrained Policy Gradient with Negative Sample Augmentation (BCPG-NSA)<n>BCPG-NSA is a fine-grained offline framework that encompasses three stages: 1) sample segmentation, 2) consensus-based step correctness assessment combining LLM and PRM judgers, and 3) policy optimization with NSA designed to effectively mine positive steps within negative samples.<n> Experimental results show that BCPG-NSA outperforms baselines on several challenging math/coding reasoning benchmarks using the same training dataset.
  arXiv  Detail & Related papers  (2025-05-20T14:16:49Z)
• Unlocking the Potential of Difficulty Prior in RL-based Multimodal Reasoning [69.64809103333839]
  We investigate how explicitly modeling problem's difficulty prior information shapes the effectiveness of reinforcement learning based fine-tuning for multimodal reasoning.<n>Our approach demonstrates significant performances across various multi-modal mathematical reasoning benchmarks with only 2K+0.6K two-stage training data.
  arXiv  Detail & Related papers  (2025-05-19T15:43:10Z)
• RL on Incorrect Synthetic Data Scales the Efficiency of LLM Math Reasoning by Eight-Fold [41.28168368547099]
  Training on model-generated synthetic data is a promising approach for finetuning LLMs, but it remains unclear when it helps or hurts. We show that training on per-step negatives can help to unlearn spurious correlations in the positive data.
  arXiv  Detail & Related papers  (2024-06-20T17:45:54Z)
• When hard negative sampling meets supervised contrastive learning [17.173114048398947]
  We introduce a new supervised contrastive learning objective, SCHaNe, which incorporates hard negative sampling during the fine-tuning phase. SchaNe outperforms the strong baseline BEiT-3 in Top-1 accuracy across various benchmarks. Our proposed objective sets a new state-of-the-art for base models on ImageNet-1k, achieving an 86.14% accuracy.
  arXiv  Detail & Related papers  (2023-08-28T20:30:10Z)
• Robust Learning with Progressive Data Expansion Against Spurious Correlation [65.83104529677234]
  We study the learning process of a two-layer nonlinear convolutional neural network in the presence of spurious features. Our analysis suggests that imbalanced data groups and easily learnable spurious features can lead to the dominance of spurious features during the learning process. We propose a new training algorithm called PDE that efficiently enhances the model's robustness for a better worst-group performance.
  arXiv  Detail & Related papers  (2023-06-08T05:44:06Z)
• Are Sample-Efficient NLP Models More Robust? [90.54786862811183]
  We investigate the relationship between sample efficiency (amount of data needed to reach a given ID accuracy) and robustness (how models fare on OOD evaluation) We find that higher sample efficiency is only correlated with better average OOD robustness on some modeling interventions and tasks, but not others. These results suggest that general-purpose methods for improving sample efficiency are unlikely to yield universal OOD robustness improvements, since such improvements are highly dataset- and task-dependent.
  arXiv  Detail & Related papers  (2022-10-12T17:54:59Z)
• Towards Robust Visual Question Answering: Making the Most of Biased Samples via Contrastive Learning [54.61762276179205]
  We propose a novel contrastive learning approach, MMBS, for building robust VQA models by Making the Most of Biased Samples. Specifically, we construct positive samples for contrastive learning by eliminating the information related to spurious correlation from the original training samples. We validate our contributions by achieving competitive performance on the OOD dataset VQA-CP v2 while preserving robust performance on the ID dataset VQA v2.
  arXiv  Detail & Related papers  (2022-10-10T11:05:21Z)
• Rethinking InfoNCE: How Many Negative Samples Do You Need? [54.146208195806636]
  We study how many negative samples are optimal for InfoNCE in different scenarios via a semi-quantitative theoretical framework. We estimate the optimal negative sampling ratio using the $K$ value that maximizes the training effectiveness function.
  arXiv  Detail & Related papers  (2021-05-27T08:38:29Z)

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.