Abstract
Efficient multimodal document question answering with explicit visual grounding, locating the precise document region that supports each answer remains an open challenge. Current approaches bifurcate into Supervised Fine-Tuning (SFT), which requires large annotated datasets and reaches optimization plateaus, and reasoning-centric Reinforcement Learning (RL), which depends on verbose intermediate traces that inflate inference token cost without clear benefit.
We introduce Perception-RFT, a training framework that applies Group Relative Policy Optimization (GRPO) to multimodal document QA, bypassing intermediate reasoning tokens to directly align visual features with structured grounding outputs. To rigorously evaluate the necessity of reasoning, we construct a reasoning variant under identical reward settings.
We find that reasoning-enabled models suppress their reasoning traces during training, converging to direct perception-based policies at the 4B parameter scale, reducing per-query inference token length by more than 60%, while reasoning-enabled RL underperforms perception-only training. Through a fine-grained analysis of Qwen3-VL-4B optimization dynamics, we confirm that SFT saturation and cold-start RL instability established in text-domain post-training extend to multimodal, and identify a previously uncharacterized Grounding Divergence: a selective trade-off between semantic robustness and geometric precision on two out-of-distribution (OOD) benchmarks (4,828 samples) under joint RL optimization.
We further show that an early SFT$ ightarrow$RL transition achieves comparable precision with 65% less training data.
Blogger's Review: The Perception-RFT framework proposed in this paper significantly enhances the efficiency of multimodal document question answering by reducing the reliance on reasoning tokens, showcasing the advantages of direct perception. This research not only provides new insights for the design of future multimodal QA systems but also opens new directions for the application of reinforcement learning in complex tasks. By optimizing strategy adjustments, the researchers effectively balance model performance and resource consumption, warranting further attention and exploration.