← 研踪 PaperTrailarXiv:2605.12374 · v5
Fill the GAP: A Granular Alignment Paradigm for Visual Reasoning in Multimodal Large Language Models
首次提交 2026-05-13 00:41
版本更新 2026-11-11 11:21
雷达观测 2026-11-13 11:11
摘要
Visual latent reasoning lets a multimodal large language model (MLLM) create intermediate visual evidence as continuous tokens, avoiding external tools or image generators. However, existing methods usually follow an output-as-input latent paradigm and yield unstable gains. We identify a feature-space mismatch behind this instability: dominant visual-latent models build on pre-norm MLLMs and reuse decoder hidden states as predicted latent inputs, even though these states occupy a substantially different norm regime from the input embeddings the model was trained to consume. This mismatch makes direct latent feedback unreliable. Motivated by this diagnosis, we propose GAP, a Granular Alignment Paradigm for visual latent modeling. GAP aligns visual latent reasoning at three levels: data-level alignment grounds latent targets with inspectable auxiliary visual supervision; feature-level alignment maps decoder outputs into input-compatible visual latents through a lightweight PCA-aligned latent head; and model-level alignment assigns latent supervision selectively to examples where the base MLLM struggles. On Qwen2.5-VL 7B, GAP achieves the best aggregate perception and reasoning scores among all compared methods and is the only evaluated latent method that improves both over the base model. Inference-time interventions further show that generated latents carry task-relevant visual signal rather than merely adding token slots. Code is available at https://github.com/Qwen-Applications/GAP.
作者备注：Accepted at NeurIPS 2026. 26 pages
观测到的版本
相邻记录可能跳过中间版本；首次发现 v2 及以上也会标为版本更新。
v5 · 2026-11-11 11:21 · 最新观测版本
Fill the GAP: A Granular Alignment Paradigm for Visual Reasoning in Multimodal Large Language Models
Visual latent reasoning lets a multimodal large language model (MLLM) create intermediate visual evidence as continuous tokens, avoiding external tools or image generators. However, existing methods usually follow an output-as-input latent paradigm and yield unstable gains. We identify a feature-space mismatch behind this instability: dominant visual-latent models build on pre-norm MLLMs and reuse decoder hidden states as predicted latent inputs, even though these states occupy a substantially different norm regime from the input embeddings the model was trained to consume. This mismatch makes direct latent feedback unreliable. Motivated by this diagnosis, we propose GAP, a Granular Alignment Paradigm for visual latent modeling. GAP aligns visual latent reasoning at three levels: data-level alignment grounds latent targets with inspectable auxiliary visual supervision; feature-level alignment maps decoder outputs into input-compatible visual latents through a lightweight PCA-aligned latent head; and model-level alignment assigns latent supervision selectively to examples where the base MLLM struggles. On Qwen2.5-VL 7B, GAP achieves the best aggregate perception and reasoning scores among all compared methods and is the only evaluated latent method that improves both over the base model. Inference-time interventions further show that generated latents carry task-relevant visual signal rather than merely adding token slots. Code is available at https://github.com/Qwen-Applications/GAP.
查看此版本解读 →查看此版本原文 ↗