CogDoc: Towards Unified thinking in Documents

TL;DR

CogDoc introduces a unified coarse-to-fine reasoning framework that mimics human cognition, improving scalability and detail in document understanding, and achieves state-of-the-art results on complex visual document benchmarks.

Contribution

Proposes CogDoc, a novel unified thinking framework combining coarse and fine reasoning, with effective post-training strategies, notably direct RL, outperforming prior methods.

Findings

Direct RL outperforms supervised fine-tuning in training.

7B model surpasses larger proprietary models on visual document benchmarks.

Unified framework improves scalability and reasoning fidelity.

Abstract

Current document reasoning paradigms are constrained by a fundamental trade-off between scalability (processing long-context documents) and fidelity (capturing fine-grained, multimodal details). To bridge this gap, we propose CogDoc, a unified coarse-to-fine thinking framework that mimics human cognitive processes: a low-resolution "Fast Reading" phase for scalable information localization,followed by a high-resolution "Focused Thinking" phase for deep reasoning. We conduct a rigorous investigation into post-training strategies for the unified thinking framework, demonstrating that a Direct Reinforcement Learning (RL) approach outperforms RL with Supervised Fine-Tuning (SFT) initialization. Specifically, we find that direct RL avoids the "policy conflict" observed in SFT. Empirically, our 7B model achieves state-of-the-art performance within its parameter class, notably surpassing significantly larger proprietary models (e.g., GPT-4o) on challenging, visually rich document benchmarks.