Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Deep Pre-Alignment for VLMs

About

Most Vision Language Models (VLMs) directly map outputs from ViT encoders to the LLM via a lightweight projector. While effective, recent analysis suggests this architecture suffers from an alignment challenge: visual features remain distant from the text space in the initial layers of the LLM, forcing the model to waste critical depth~\cite{zhang-etal-2024-investigating,artzy-schwartz-2024-attend} on superficial modality alignment rather than deep understanding and complex reasoning. In this work, we propose Deep Pre-Alignment (DPA), a novel architecture that replaces the standard ViT encoder with a small VLM as perceiver, ensuring visual features are deeply aligned with the text space of the target large language model. Comprehensive experiments demonstrate the effectiveness of DPA. On the 4B parameter scale, DPA outperforms baselines by 1.9 points across 8 multimodal benchmarks, with gains widening to 3.0 points at the 32B scale. Moreover, by offloading alignment to the perceiver, DPA achieves a 32.9\% reduction in language capability forgetting over 3 text benchmarks. We further demonstrate that these gains are consistent across different LLM families including Qwen3 and LLaMA 3.2, highlighting the generality of our approach. Beyond performance, DPA also offers a seamless upgrade path for current VLM development, requiring only a modular replacement for the visual encoder with marginal computation overhead.

Tianyu Yu, Kechen Fang, Zihao Wan, Kaidong Zhang, Yicheng Zhang, Jun Song, Bo Zheng, Yuan Yao• 2026

Related benchmarks

TaskDatasetResultRank
Visual UnderstandingMM-Vet
MM-Vet Score56
167
General image understandingMMStar
Accuracy59.1
58
Visual PerceptionAI2D
Accuracy79
47
Language UnderstandingMMLU-Redux
Accuracy80.6
29
Multimodal ReasoningMMMU
MMMU Score57.1
27
Visual PerceptionOCRBench
Score76.8
22
Holistic EvaluationCombined Suite General Reasoning Perception Text
Text Average58.1
13
Science ReasoningGPQA Diamond
GPQA Diamond35.4
13
General Visual UnderstandingSeedBench-2-Plus
SeedBench2+ Score67.7
11
Multimodal ReasoningMathVista
MathVista Score63.7
11
Showing 10 of 11 rows

Other info

Follow for update