Research Papers 论文研究 5h ago Updated 59m ago 更新于 59分钟前 46

Towards Understanding Pause Token Fine-Tuning Dynamics: A Mode Retention Perspective 迈向理解暂停令牌微调动力学:一种模式保持视角

Pause tokens improve LLM reasoning not just through computational expressivity but by reshaping fine-tuning training dynamics, specifically the retention-adaptation trade-off Two controlled pilots reveal distinct asymmetries: masked pauses overwrite previously-learned distributions ~4x less (mode retention), and boundary-adjacent tokens encode substantially more downstream-step information (non-myopic compression) Masked Boundary Pause (MBP) formalizes a training rule placing pause tokens at rea 提出Masked Boundary Pause (MBP)训练方法,通过在推理步骤边界插入pause token并mask其loss,改善LLM推理能力 两个控制实验揭示pause token的两种动态不对称性:H1模式保留(masked pauses覆盖先前学习分布少4倍)和H2非短视压缩(边界token编码更多下游步骤信息) 在1B-8B Qwen和Llama模型上验证,MBP在数学推理上提升最高6分、代码提升2.5分,同时保持通用语言能力 将pause token重新定义为训练动态干预手段(解决保留-适应权衡),而非仅作为推理时计算设备 该模式保留策略可扩展至GRPO训练框架,进一步验证

60
Hot 热度
72
Quality 质量
65
Impact 影响力

Analysis 深度分析

TL;DR

  • Pause tokens improve LLM reasoning not just through computational expressivity but by reshaping fine-tuning training dynamics, specifically the retention-adaptation trade-off
  • Two controlled pilots reveal distinct asymmetries: masked pauses overwrite previously-learned distributions ~4x less (mode retention), and boundary-adjacent tokens encode substantially more downstream-step information (non-myopic compression)
  • Masked Boundary Pause (MBP) formalizes a training rule placing pause tokens at reasoning-step boundaries with masked loss, consistently improving reasoning across 1B-8B Qwen and Llama models
  • MBP achieves gains of up to 6 points on math benchmarks and 2.5 points on code benchmarks while preserving general language understanding abilities
  • The mode-preserving strategy extends gains to GRPO (Group Relative Policy Optimization), broadening its applicability beyond supervised fine-tuning

Why It Matters

This work reframes pause tokens from being viewed as mere inference-time computation devices to understanding them as active training-dynamics interventions that mitigate catastrophic forgetting during fine-tuning. For AI practitioners, this provides both a theoretical foundation and a practical method (MBP) to improve reasoning capabilities without sacrificing general language understanding—a persistent challenge in LLM fine-tuning pipelines.

Technical Details

  • Masked Boundary Pause (MBP): A training rule where special pause tokens are inserted at reasoning-step boundaries and their loss is masked during backpropagation, preventing gradient updates from overwriting previously learned distributions at those positions
  • Mode Retention (H1): On a synthetic continual-learning task, masked pauses reduce overwriting of previously-learned distributions by approximately 4x compared to unmasked counterparts at matched final adaptation levels
  • Non-Myopic Compression (H2): On a synthetic math-reasoning probe, tokens adjacent to pause boundaries learn to encode substantially more information about downstream reasoning steps, suggesting pause tokens enable longer-horizon information flow
  • Empirical validation across 1B-8B Qwen and Llama models demonstrates consistent reasoning improvements: up to +6 points on math benchmarks and +2.5 points on code benchmarks, with no degradation in general language understanding metrics
  • GRPO extension: The mode-preserving strategy is shown to be compatible with reinforcement learning-based alignment methods, extending gains beyond supervised fine-tuning to group-relative policy optimization

Industry Insight

  • The retention-adaptation trade-off framework offers a new lens for diagnosing and addressing catastrophic forgetting in fine-tuning—practitioners should consider loss-masking strategies at architectural decision points, not just regularization techniques
  • MBP's compatibility with GRPO suggests pause-token interventions can be integrated into full RLHF-style pipelines, making them viable for production reasoning models without requiring separate fine-tuning stages
  • The finding that boundary-adjacent tokens encode more downstream information implies that strategic token placement (not just token count) matters for reasoning performance—future architectures should optimize pause token positioning rather than treating them as uniform insertions

TL;DR

  • 提出Masked Boundary Pause (MBP)训练方法,通过在推理步骤边界插入pause token并mask其loss,改善LLM推理能力
  • 两个控制实验揭示pause token的两种动态不对称性:H1模式保留(masked pauses覆盖先前学习分布少4倍)和H2非短视压缩(边界token编码更多下游步骤信息)
  • 在1B-8B Qwen和Llama模型上验证,MBP在数学推理上提升最高6分、代码提升2.5分,同时保持通用语言能力
  • 将pause token重新定义为训练动态干预手段(解决保留-适应权衡),而非仅作为推理时计算设备
  • 该模式保留策略可扩展至GRPO训练框架,进一步验证其泛化价值

为什么值得看

本文首次从训练动态视角系统研究pause token机制,突破了以往仅从计算表达能力解释其效果的局限,为LLM推理优化提供了新的理论框架。研究结果对实际微调实践具有直接指导意义,特别是在平衡推理能力提升与通用能力保持方面。

技术解析

  • MBP训练规则:在推理步骤边界放置pause token,并mask其loss,使模型在关键决策点保留已有知识模式,同时允许推理能力逐步适应
  • H1模式保留实验:在合成持续学习任务上,masked pauses在匹配的最终适应阶段约少覆盖4倍先前学习的分布,证明pause token有助于知识保留
  • H2非短视压缩实验:在合成数学推理探针上,边界相邻token编码了显著更多的下游步骤信息,表明pause token促进了前瞻性信息压缩
  • 实验规模:覆盖1B-8B参数的Qwen和Llama系列模型,在数学和代码推理基准上验证,同时评估通用语言理解能力保持情况
  • GRPO扩展:证明MBP的模式保留策略可延伸至GRPO训练框架,进一步验证其有效性和泛化性

行业启示

  • 微调策略优化:MBP为LLM推理能力微调提供了可落地的训练干预方案,帮助工程师在提升专项能力时避免灾难性遗忘
  • 推理token设计范式转变:研究推动行业从"推理时计算设备"转向"训练动态干预工具"的认知,可能影响未来pause token类方法的设计方向
  • 持续学习平衡:模式保留机制为解决LLM持续学习中的适应-保留权衡提供了新思路,对需要多任务迭代的工业场景具有参考价值

Disclaimer: The above content is generated by AI and is for reference only. 免责声明:以上内容由 AI 生成,仅供参考。

LLM 大模型 Fine-tuning 微调 Research 科学研究 Training 训练