提出新剪枝方法，让多模态模型在瘦身时仍保持推理能力。
MuCRASP: Multimodal Chain-of-thought Reasoning aware Structured Pruning
- 针对推理关键点设计结构化剪枝，保留思维链轨迹
- 30%剪枝下推理得分8.87，优于基线1.55分
- 适合需要高效部署复杂推理多模态模型的研究者
视觉语言模型（VLMs）依赖思维链（CoT）推理解决复杂多模态任务，但参数量大导致部署成本高。结构化剪枝是自然解决方案，但现有方法无法保留CoT推理准确性。我们发现两大原因：(1) CoT一致性依赖生成轨迹中的稀疏转折点（枢轴标记），而现有剪枝方法对此无感知；(2) 针对单模态大模型的剪枝方法未考虑视觉与文本模态间的激活分布差异。为此，我们提出MuCRASP，一种面向推理关键组件的结构化剪枝框架，在全局参数预算下保留跨模态对齐性并考虑层间敏感度。在四个VLMs上三个推理基准测试中，MuCRASP在压缩率提升时仍稳定保持推理质量。在Qwen2.5-VL-7B上30%剪枝时，其LLM-as-a-Judge得分为8.87，显著优于最强基线的7.32。此外，其推理一致性可维持至50%剪枝，且困惑度下降更小。
原文摘要 · Abstract (English)
Vision-language models (VLMs) increasingly rely on chain-of-thought (CoT) reasoning to solve complex multimodal tasks, but their large parameter sizes make deployment expensive. Structured pruning offers a natural solution; however, existing methods fail to preserve CoT reasoning accuracy in VLMs. We identify two key reasons: (1) CoT consistency depends on sparse transition points (pivot tokens) in the generation trajectory, while existing pruning methods are CoT-agnostic; and (2) pruning methods designed for unimodal LLMs do not account for activation-distribution differences across visual and textual modalities. Motivated by these observations, we propose MuCRASP, a structured pruning framework that targets reasoning-critical components while preserving cross-modal alignment and accounting for layer-wise sensitivity under a global parameter budget. Experiments on four VLMs across three reasoning benchmarks show that MuCRASP consistently preserves reasoning quality under increasing compression. At 30% pruning on Qwen2.5-VL-7B, MuCRASP achieves an LLM-as-a-Judge score of 8.87 versus 7.32 for the strongest baseline on physical reasoning tasks. Furthermore, MuCRASP maintains high reasoning consistency up to 50% pruning, significantly outperforming prior pruning approaches while exhibiting lower perplexity degradation.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。