Internalized Reasoning for Long-Context Visual Document Understanding
Visual long-document understanding is critical for enterprise, legal, and scientific applications, yet the best performing open recipes have not explored reasoning, a capability which has driven leaps in math and code performance. We introduce a synthetic data pipeline for reasoning in long-document understanding that generates thinking traces by scoring each page for question relevance, extracting textual evidence and ordering it from most to least relevant. We apply SFT to the resulting traces within \texttt{<think>} tags, gated by a \texttt{<cot>} control token, and the resulting reasoning capability is internalized via low-strength model merging. We study Qwen3 VL 32B and Mistral Small 3.1 24B. With Qwen3 VL, we achieve 58.3 on MMLongBenchDoc, surpassing the 7$\times$ larger Qwen3 VL 235B A22B (57.0). With Mistral, we show that synthetic reasoning outperforms distillation from the Thinking version's traces by 3.8 points on MMLBD-C, and internalized reasoning exhibits 12.4$\times$ fewer mean output tokens compared to explicit reasoning. We release our pipeline for reproducibility and further exploration.
Code (0)
등록된 구현이 없습니다.
Similar Papers 제목 키워드 기반
TInR: Exploring Tool-Internalized Reasoning in Large Language Models
Tool-Integrated Reasoning (TIR) has emerged as a promising direction by extending Large Language Models' (LLMs) capabilities with external tools during reasoning. Existing TIR methods typically rely on external tool docu…
Reinforcement LearningSTORM: Internalized Modeling for Spatial-Temporal Reasoning in Video-Language Models
Many video reasoning tasks require tracking motion, temporal order, and evolving visual states across frames. Existing methods built on large vision-language models (LVLMs) often address this challenge by externalizing r…
Visual ReasoningBeyond Visual CoT: Internalized Visual Thinking for Proactive Video Reasoning
Multimodal large language models increasingly use visual chain-of-thought (Visual CoT) to reason about spatial, temporal, and embodied environments. By generating intermediate reasoning images, Visual CoT provides an int…
GRC: Unifying Reasoning-Driven Generation, Retrieval and Compression
Text embedding and generative tasks are usually trained separately based on large language models (LLMs) nowadays. This causes a large amount of training cost and deployment effort. Context compression is also a challeng…
Continual LearningInSight-doc: Agentic Visual Perception for Long-Document Understanding
Long-document understanding often requires reasoning over many visually rich pages, making inference costly and prone to context rot. In this work, we propose InSight-doc, an agentic visual perception framework that trea…