As vision-language models (VLMs) increasingly succeed in multimodal reasoning benchmarks, their visual evidence often becomes unstable once it enters the language stack, weakening evidence-grounded reasoning.
To understand this fragility, we examine the internal dynamics of VLMs through a mechanistic lens and uncover a stable three-stage redistribution of multimodal attention focus across depth: an early question-conditioned organization, a critical middle visual-dominant relay, and a late return to answer formation.
We operationalize the middle phase as the Visual Relay Window (VRW) and show that its geometry varies with task demand, is causally tied to grounded generation, and distinguishes unsupported answers from stronger reasoning trajectories.
Guided by this internal rhythm, we propose TRACE, a task-adaptive inference-time control framework with lightweight trained modules. It reshapes relay allocation during prefill and preserves assembled visual support after handoff during decoding.
Across four open-weight VLM backbones and seven benchmarks, TRACE delivers large gains on grounding-sensitive settings, improving them by an average of 4.33 points and up to 6.6 points, while also improving reasoning-heavy tasks.
These results show that explicitly controlling multimodal focus across depth offers a unified and effective mechanism for strengthening evidence-grounded multimodal reasoning.
Blogger's Review: The TRACE framework proposed in this paper provides a fresh perspective for vision-language models by enhancing reasoning capabilities through precise control of multimodal focus, especially when handling complex visual evidence. Its effectiveness has been validated across multiple benchmark tests, demonstrating its potential in practical applications. Future research could explore the applicability of TRACE in other domains and how to further optimize its performance.