Beyond Next-Token Alignment: Distilling Multimodal Large Language Models via Token Interactions

• URL: http://arxiv.org/abs/2602.09483v1
• Date: Tue, 10 Feb 2026 07:26:56 GMT
• Title: Beyond Next-Token Alignment: Distilling Multimodal Large Language Models via Token Interactions
• Authors: Lin Chen, Xiaoke Zhao, Kun Ding, Weiwei Feng, Changtao Miao, Zili Wang, Wenxuan Guo, Ying Wang, Kaiyuan Zheng, Bo Zhang, Zhe Li, Shiming Xiang,
• Abstract summary: We introduce Align-TI, a novel KD framework designed from the perspective of Token Interactions.<n>Our approach is motivated by the insight that MLLMs rely on two primary interactions: vision-instruction token interactions to extract relevant visual information, and intra-response token interactions for coherent generation.
• Score: 33.54873330567528
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Multimodal Large Language Models (MLLMs) demonstrate impressive cross-modal capabilities, yet their substantial size poses significant deployment challenges. Knowledge distillation (KD) is a promising solution for compressing these models, but existing methods primarily rely on static next-token alignment, neglecting the dynamic token interactions, which embed essential capabilities for multimodal understanding and generation. To this end, we introduce Align-TI, a novel KD framework designed from the perspective of Token Interactions. Our approach is motivated by the insight that MLLMs rely on two primary interactions: vision-instruction token interactions to extract relevant visual information, and intra-response token interactions for coherent generation. Accordingly, Align-TI introduces two components: IVA enables the student model to imitate the teacher's instruction-relevant visual information extract capability by aligning on salient visual regions. TPA captures the teacher's dynamic generative logic by aligning the sequential token-to-token transition probabilities. Extensive experiments demonstrate Align-TI's superiority. Notably, our approach achieves $2.6\%$ relative improvement over Vanilla KD, and our distilled Align-TI-2B even outperforms LLaVA-1.5-7B (a much larger MLLM) by $7.0\%$, establishing a new state-of-the-art distillation framework for training parameter-efficient MLLMs. Code is available at https://github.com/lchen1019/Align-TI.

Related papers

• Reconstructing Content via Collaborative Attention to Improve Multimodal Embedding Quality [59.651410243721045]
  CoCoA is a Content reconstruction pre-training paradigm based on Collaborative Attention for multimodal embedding optimization.<n>We introduce an EOS-based reconstruction task, encouraging the model to reconstruct input from the corresponding EOS> embeddings.<n>Experiments on MMEB-V1 demonstrate that CoCoA built upon Qwen2-VL and Qwen2.5-VL significantly improves embedding quality.
  arXiv  Detail & Related papers  (2026-03-02T05:34:45Z)
• MoCa: Modality-aware Continual Pre-training Makes Better Bidirectional Multimodal Embeddings [75.0617088717528]
  MoCa is a framework for transforming pre-trained VLM backbones into effective bidirectional embedding models.<n>MoCa consistently improves performance across MMEB and ViDoRe-v2 benchmarks, achieving new state-of-the-art results.
  arXiv  Detail & Related papers  (2025-06-29T06:41:00Z)
• M$^2$IV: Towards Efficient and Fine-grained Multimodal In-Context Learning via Representation Engineering [21.75002972755496]
  Multimodal in-context learning (ICL) equips Large Vision-language Models (LVLMs) with the ability to adapt to new tasks via multiple user-provided demonstrations.<n>However, its effectiveness is constrained by the token-intensive nature of multimodal inputs and the complexity of cross-modal few-shot reasoning.<n>We propose textbfM$2$IV, a novel representation engineering approach that replaces explicit token-level demonstrations with a set of learnable Multimodal In-context Vectors.
  arXiv  Detail & Related papers  (2025-04-06T22:02:21Z)
• Knowledge-Decoupled Synergetic Learning: An MLLM based Collaborative Approach to Few-shot Multimodal Dialogue Intention Recognition [17.790383360652704]
  Training for few-shot multimodal dialogue intention recognition involves two interconnected tasks.<n>This phenomenon is attributed to knowledge interference stemming from the superposition of weight matrix updates during the training process.<n>We propose Knowledge-Decoupled Synergetic Learning, which transforms knowledge into interpretable rules, while applying the post-training of larger models.
  arXiv  Detail & Related papers  (2025-03-06T08:28:44Z)
• ADEM-VL: Adaptive and Embedded Fusion for Efficient Vision-Language Tuning [38.26304604660713]
  ADEM-VL is an efficient vision-language method that tunes models based on pretrained large language models. Our framework surpasses existing methods by an average accuracy of 0.77% on ScienceQA dataset.
  arXiv  Detail & Related papers  (2024-10-23T11:31:06Z)
• Interactive DualChecker for Mitigating Hallucinations in Distilling Large Language Models [7.632217365130212]
  Large Language Models (LLMs) have demonstrated exceptional capabilities across various machine learning (ML) tasks. These models can produce hallucinations, particularly in domains with incomplete knowledge. We introduce DualChecker, an innovative framework designed to mitigate hallucinations and improve the performance of both teacher and student models.
  arXiv  Detail & Related papers  (2024-08-22T12:04:04Z)
• LLAVADI: What Matters For Multimodal Large Language Models Distillation [77.73964744238519]
  In this work, we do not propose a new efficient model structure or train small-scale MLLMs from scratch. Our studies involve training strategies, model choices, and distillation algorithms in the knowledge distillation process. By evaluating different benchmarks and proper strategy, even a 2.7B small-scale model can perform on par with larger models with 7B or 13B parameters.
  arXiv  Detail & Related papers  (2024-07-28T06:10:47Z)
• NoteLLM-2: Multimodal Large Representation Models for Recommendation [71.87790090964734]
  Large Language Models (LLMs) have demonstrated exceptional proficiency in text understanding and embedding tasks.<n>Their potential in multimodal representation, particularly for item-to-item (I2I) recommendations, remains underexplored.<n>We propose an end-to-end fine-tuning method that customizes the integration of any existing LLMs and vision encoders for efficient multimodal representation.
  arXiv  Detail & Related papers  (2024-05-27T03:24:01Z)
• LION : Empowering Multimodal Large Language Model with Dual-Level Visual Knowledge [58.82222646803248]
  Multimodal Large Language Models (MLLMs) have endowed LLMs with the ability to perceive and understand multi-modal signals. Most of the existing MLLMs mainly adopt vision encoders pretrained on coarsely aligned image-text pairs, leading to insufficient extraction and reasoning of visual knowledge. We propose a dual-Level vIsual knedgeOwl eNhanced Multimodal Large Language Model (LION), which empowers the MLLM by injecting visual knowledge in two levels.
  arXiv  Detail & Related papers  (2023-11-20T15:56:44Z)
• Pink: Unveiling the Power of Referential Comprehension for Multi-modal LLMs [49.88461345825586]
  This paper proposes a new framework to enhance the fine-grained image understanding abilities of MLLMs. We present a new method for constructing the instruction tuning dataset at a low cost by leveraging annotations in existing datasets. We show that our model exhibits a 5.2% accuracy improvement over Qwen-VL and surpasses the accuracy of Kosmos-2 by 24.7%.
  arXiv  Detail & Related papers  (2023-10-01T05:53:15Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.