TokenSwap introduces a method to benchmark and reduce the modality gap in multimodal large language models (MLLMs) by interleaving semantically aligned images with text tokens. By transforming text-only benchmarks like MMLU into image‑interleaved versions (TokenSwap‑Bench), the authors reveal a pervasive performance drop—averaging 19.6%—across 42 MLLMs, with reasoning models showing smaller gaps. Training with TokenSwap mitigates this gap while maintaining strong text‑only and vision‑language performance.
By Andong Hua, Colton Bishop, Igor Mordatch, Arian Hosseini, Jindong Gu, Aleksandra Faust, Rebecca Roelofs, Yao Qin
The paper introduces Vision-Free Adaptation (VFA), a method that separates multilingual language enhancement from visual alignment in multimodal large language models. VFA fine‑tunes a base LLM on multilingual text to create a multilingual task vector, which is then merged with the vision‑aligned task vector of an existing MLLM. Experiments on five MLLMs and six multilingual benchmarks show consistent gains while preserving multimodal and text‑only performance, and using less than 2% of text data narrows the performance gap to fully multimodal‑trained models.
By Yixia Li, Yaqing Shi, Zhiwen Ruan, Dongdong Zhang, Lingjie Jiang, Shaohan Huang, Yun Chen, Guanhua Chen, Furu Wei
arXiv:2602. 07026v3 Announce Type: replace-cross Abstract: Despite the success of multimodal contrastive learning in aligning visual and linguistic representations, a persistent geometric anomaly, the Modality Gap, remains: embeddings of distinct modalities expressing identical semantics occupy systematically offset regions.
By Xiaomin Yu, Yi Xin, Yuhui Zhang, Wenjie Zhang, Chonghan Liu, Hanzhen Zhao, Chen Liu, Xiaoxing Hu, Ziyue Qiao, Hao Tang, Xiaobin Hu, Chengwei Qin, Hui Xiong, Yu Qiao, Shuicheng Yan
Wild test-time adaptation (WTTA) updates a source model online under small test batches, concurrent distribution shifts, and time-varying class imbalance. Most WTTA methods derive their adaptation sig...
The paper introduces selective adaptation for vision‑language models, questioning whether test‑time adaptation (TTA) should always be applied. By analyzing per‑sample predictions before and after adaptation, the authors find that many adaptations are negligible or even harmful, flipping correct predictions. They propose Cross‑Augmentation Similarity (CAS), which skips adaptation when predictions across augmented views are highly similar, achieving comparable or better accuracy while reducing adaptation by up to 85%.
By Siru Jiang, Yuwei Liang, Jian Liang, Ran He, Tieniu Tan
The paper introduces MASA, a method for Wild Test-Time Adaptation that uses a frozen multimodal large language model to provide structured semantic anchors, thereby avoiding the self-referential loop common in existing WTTA techniques. MASA selects a small, diverse set of reliability-ranked anchors, encodes their descriptions, propagates them to nearby test samples, and stores this visual‑semantic information in an online prototype memory. The stored descriptors enable lightweight adaptation of normalization parameters, and MASA is evaluated on the WTTA ImageNet‑C benchmark with ResNet and ViT backbones under limited‑batch, mixed‑domain, and imbalanced‑label‑shift scenarios.
By Zhenbin Wang, Lei Zhang, Lituan Wang, Yan Wang, Zhao Zhang, Wei Huang
arXiv:2605. 18160v2 Announce Type: replace-cross Abstract: In recent years, multimodal large language models (MLLMs) have achieved remarkable progress, primarily attributed to effective paradigms for integrating visual and textual information.
By Xinpeng Dong, Min Zhang, Kairong Han, Xu Tan, Fei Wu, Kun Kuang
Multi-modal Large Language Models (MLLMs) achieve strong performance on vision-language tasks, but incorporating visual inputs through a vision encoder (e. g.
arXiv:2607. 26596v1 Announce Type: cross Abstract: Multimodal large language models (MLLMs) have demonstrated remarkable capabilities by integrating visual and textual understanding within a unified transformer architecture.
By Mingkuan Feng, Zhengqi Wen, Jianhua Tao
arXiv:2607. 09450v1 Announce Type: cross Abstract: Pre-trained Vision-Language Models (VLMs) such as CLIP achieve strong zero-shot generalization, but their performance degrades sharply under adversarial perturbations.
By Xingyu Zhu, Huanshen Wu, Shuo Wang, Beier Zhu, Jiannan Ge, Jiaheng Zhang, Long Chen
arXiv:2606. 14299v1 Announce Type: cross Abstract: Vision-Language Models (VLMs) such as CLIP have become a standard backbone for open-vocabulary recognition, yet their zero-shot predictions remain vulnerable to distribution shifts encountered at deployment.
By Jiazhen Huang, Xiao Chen, Zhiming Liu, Yaru Sun, Jingyan Jiang, Zhi Wang
arXiv:2602.06652v2 Announce Type: replace
Abstract: The robustness of Vision Language Models (VLMs) is commonly assessed through output-level invariance, implicitly assuming that stable predictions r...
By Farooq Ahmad Wani, Alessandro Suglia, Rohit Saxena, Aryo Pradipta Gema, Wai-Chung Kwan, Fazl Barez, Maria Sofia Bucarelli, Fabrizio Silvestri, Pasquale Minervini