MCPO: Modality-Contrastive Preference Optimization for Multimodal Chain-of-Thought Compression

• URL: http://arxiv.org/abs/2609.04947v2
• Date: Tue, 08 Sep 2026 03:11:34 GMT
• Title: MCPO: Modality-Contrastive Preference Optimization for Multimodal Chain-of-Thought Compression
• Abstract summary: We propose a two-stage length-compression method that requires fewer than 900 training samples.<n>In the compression stage, we introduce a step-level Normalized Cross-Modal Mutual Information pruning algorithm.<n>In the alignment stage, we use an asymmetric multimodal length-controlled preference loss.<n>Experiments show that our method can reduce CoT length by up to 69.5% and achieve up to 3.34x end-to-end inference speedup.
• Score: 40.580984703096945
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Recently, multimodal large-scale reasoning models have demonstrated remarkable capabilities in solving complex tasks through long Chains-of-Thought (M-CoT). However, excessively long reasoning trajectories incur substantial computational costs and significant KV-cache pressure. Existing CoT compression and alignment paradigms mainly rely on static rules or single-dimensional preferences, lacking fine-grained cross-modal constraints; as a result, they are prone to inducing visual laziness and hallucinatory reasoning. To address these issues, we propose Modality-Contrastive Preference Optimization (MCPO), a highly sample-efficient two-stage length-compression method that requires fewer than 900 training samples. In the compression stage, we introduce a step-level Normalized Cross-Modal Mutual Information (NCMI) pruning algorithm, which automatically identifies and removes visual-independent reasoning steps by comparing the reasoning discrepancies between with-image and no-image contexts. This significantly reduces redundancy and hallucinatory content in the reasoning chains. In the alignment stage, the model first undergoes supervised fine-tuning to achieve domain-adaptive initialization, followed by optimization using an asymmetric multimodal length-controlled preference loss. This objective adopts a highly nonlinear odds-ratio formulation that provides steep gradients in the with-image context to reinforce length constraints for preferred trajectories, while applying a scaled, flat-gradient linear difference in the no-image context to maintain modality consistency, thereby achieving stable cross-modal preference alignment. Extensive experiments on mainstream base models such as Qwen3-VL-Thinking show that our method can reduce CoT length by up to 69.5% and achieve up to 3.34x end-to-end inference speedup while preserving original accuracy.

Related papers

• Reasoning Matters: Mitigate Hallucination in Multimodal Large Reasoning Models via Reasoning-Conditioned Preference Optimization [59.20570719781289]
  Multimodal Large Reasoning Models introduce the reasoning paradigm, demonstrating strong capabilities on complex vision-language tasks.<n>Existing training-based methods typically mitigate hallucinations through response-level direct preference optimization (DPO), where the Chain-of-Thought (CoT) and the final answer are treated as a monolithic output and optimized jointly.<n>We derive Reasoning-Conditioned Direct Preference Optimization (RC-DPO), which models the CoT as a condition for answer generation and contrasts the preference for the same preferred answer under different CoT conditions, promoting answer-supportive reasoning chain alignment.
  arXiv  Detail & Related papers  (2026-05-27T03:27:23Z)
• Deterministic Differentiable Structured Pruning for Large Language Models [37.33389749907146]
  Structured pruning reduces LLM inference cost by removing low-importance architectural components.<n>We propose Deterministic Differentiable Pruning (DDP), a mask-only optimization method.<n>Compared with prior approaches, DDP offers greater expressiveness, reduced train--test mismatch, and faster convergence.
  arXiv  Detail & Related papers  (2026-03-09T07:59:17Z)
• Towards Efficient Large Language Reasoning Models via Extreme-Ratio Chain-of-Thought Compression [55.63153956934198]
  Chain-of-Thought (CoT) reasoning successfully enhances the reasoning capabilities of Large Language Models (LLMs)<n>Existing CoT compression methods often suffer from a critical loss of logical fidelity at high compression ratios.<n>We propose a novel EXTreme-RAtio Chain-of-Thought Compression framework, termed Extra-CoT, which aggressively reduces the token budget while preserving answer accuracy.
  arXiv  Detail & Related papers  (2026-02-09T06:57:15Z)
• Training-free Context-adaptive Attention for Efficient Long Context Modeling [57.703159205740185]
  Training-free Context-adaptive Attention (TCA-Attention) is a training-free sparse attention mechanism that selectively attends to only the informative tokens for efficient long-context inference.<n>TCA-Attention achieves a 2.8$times$ speedup and reduces KV cache by 61% at 128K context length while maintaining performance comparable to full attention.
  arXiv  Detail & Related papers  (2025-12-10T01:54:57Z)
• Tuning-Free Structured Sparse Recovery of Multiple Measurement Vectors using Implicit Regularization [13.378211527081582]
  We introduce a tuning-free framework to recover sparse signals in multiple measurement vectors.<n>We show that the optimization dynamics exhibit a "momentum-like" effect, causing the norms of rows in the true support to grow significantly faster than others.
  arXiv  Detail & Related papers  (2025-12-03T02:53:11Z)
• T3: Test-Time Model Merging in VLMs for Zero-Shot Medical Imaging Analysis [15.624549727053475]
  Existing model-merging techniques fail to deliver consistent gains across diverse medical modalities.<n>We introduce Test-Time Task adaptive merging (T3), a backpropagation-free framework that computes per-sample coefficients.<n>We present a rigorous cross-evaluation protocol spanning in-domain, base-to-novel, and corruptions across four modalities.
  arXiv  Detail & Related papers  (2025-10-31T08:05:40Z)
• MPQ-DMv2: Flexible Residual Mixed Precision Quantization for Low-Bit Diffusion Models with Temporal Distillation [74.34220141721231]
  We present MPQ-DMv2, an improved textbfMixed textbfPrecision textbfQuantization framework for extremely low-bit textbfDiffusion textbfModels.
  arXiv  Detail & Related papers  (2025-07-06T08:16:50Z)
• Fractured Chain-of-Thought Reasoning [61.647243580650446]
  We introduce Fractured Sampling, a unified inference-time strategy that interpolates between full CoT and solution-only sampling.<n>We show that Fractured Sampling consistently achieves superior accuracy-cost trade-offs, yielding steep log-linear scaling gains in Pass@k versus token budget.
  arXiv  Detail & Related papers  (2025-05-19T11:30:41Z)

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.