Fine-tuning MLLMs Without Forgetting Is Easier Than You Think

• URL: http://arxiv.org/abs/2603.14493v1
• Date: Sun, 15 Mar 2026 17:16:19 GMT
• Title: Fine-tuning MLLMs Without Forgetting Is Easier Than You Think
• Abstract summary: We design a 2x2 experimental framework to assess model performance across in-distribution and out-of-distribution image and text inputs.<n>Our results show that appropriate regularization, such as constraining the number of trainable parameters or adopting a low learning rate, effectively prevents forgetting when dealing with out-of-distribution images.<n>We attribute this forgetting as task-specific overfitting and address this issue by introducing a data-hybrid training strategy.
• Score: 72.59321247529975
• License: http://creativecommons.org/licenses/by-nc-sa/4.0/
• Abstract: The paper demonstrate that simple adjustments of the fine-tuning recipes of multimodal large language models (MLLM) are sufficient to mitigate catastrophic forgetting. On visual question answering, we design a 2x2 experimental framework to assess model performance across in-distribution and out-of-distribution image and text inputs. Our results show that appropriate regularization, such as constraining the number of trainable parameters or adopting a low learning rate, effectively prevents forgetting when dealing with out-of-distribution images. However, we uncover a distinct form of forgetting in settings with in-distribution images and out-of-distribution text. We attribute this forgetting as task-specific overfitting and address this issue by introducing a data-hybrid training strategy that combines datasets and tasks. Finally, we demonstrate that this approach naturally extends to continual learning, outperforming existing methods with complex auxiliary mechanisms. In general, our findings challenge the prevailing assumptions by highlighting the inherent robustness of MLLMs and providing practical guidelines for adapting them while preserving their general capabilities.

Related papers

• Focus When Necessary: Adaptive Routing and Collaborative Grounding for Training-Free Visual Grounding [56.38477453373368]
  LazyMCoT is a dynamic and training-free framework that adaptively allocates visual grounding efforts based on sample difficulty.<n>LazyMCoT rivals training-based approaches by simultaneously improving reasoning accuracy and reducing average inference latency.
  arXiv  Detail & Related papers  (2026-06-15T03:17:44Z)
• Modest-Align: Data-Efficient Alignment for Vision-Language Models [67.48633659305592]
  Cross-modal alignment models often suffer from overconfidence and degraded performance when operating in resource-constrained settings.<n>We propose Modest-Align, a lightweight alignment framework designed for robustness and efficiency.<n>Our method offers a practical and scalable solution for cross-modal alignment in real-world, low-resource scenarios.
  arXiv  Detail & Related papers  (2025-10-24T16:11:10Z)
• Representing LLMs in Prompt Semantic Task Space [0.1784233255402269]
  Large language models (LLMs) achieve impressive results over various tasks.<n>Identifying the best-performing LLM for a given task is a significant challenge.<n>This work presents an efficient, training-free approach to representing LLMs as linear operators within the prompts' semantic task space.
  arXiv  Detail & Related papers  (2025-09-26T15:48:10Z)
• A Visual Leap in CLIP Compositionality Reasoning through Generation of Counterfactual Sets [26.167194142428475]
  Vision-language models (VLMs) often struggle with compositional reasoning due to insufficient high-quality image-text data.<n>We propose a novel block-based diffusion approach that automatically generates counterfactual datasets without manual annotation.<n>Our approach achieves state-of-the-art results across multiple benchmarks while using substantially less training data than existing methods.
  arXiv  Detail & Related papers  (2025-07-07T06:47:10Z)
• Diffusion Augmented Retrieval: A Training-Free Approach to Interactive Text-to-Image Retrieval [7.439049772394586]
  Diffusion Augmented Retrieval (DAR) is a framework that generates multiple intermediate representations via dialogue refinements and DMs.<n>DAR results on par with finetuned I-TIR models, yet without incurring their tuning overhead.
  arXiv  Detail & Related papers  (2025-01-26T03:29:18Z)
• Words Matter: Leveraging Individual Text Embeddings for Code Generation in CLIP Test-Time Adaptation [21.20806568508201]
  We show how to leverage class text information to mitigate distribution drifts encountered by vision-language models (VLMs) during test-time inference.<n>We propose to generate pseudo-labels for the test-time samples by exploiting generic class text embeddings as fixed centroids of a label assignment problem.<n>Experiments on multiple popular test-time adaptation benchmarks presenting diverse complexity empirically show the superiority of CLIP-OT.
  arXiv  Detail & Related papers  (2024-11-26T00:15:37Z)
• Distributionally robust self-supervised learning for tabular data [4.172010719137041]
  Learning robust representation in presence of error slices is challenging, due to high cardinality features and the complexity of constructing error sets.<n>Traditional robust representation learning methods are largely focused on improving worst group performance in supervised setting in computer vision.<n>Our approach utilizes an encoder-decoder model trained with Masked Language Modeling (MLM) loss to learn robust latent representations.
  arXiv  Detail & Related papers  (2024-10-11T04:23:56Z)
• Debiasing Multimodal Large Language Models via Penalization of Language Priors [38.97645845493758]
  Multimodal Large Language Models (MLLMs) have become indispensable tools in computer vision and natural language processing.<n>Despite their advancements, our investigation reveals a noteworthy bias: the generated content is often driven more by the inherent priors of the underlying Large Language Models (LLMs) than by the input image.<n>We propose two simple, training-free strategies to rectify these biases and redirect the model's focus toward visual information.
  arXiv  Detail & Related papers  (2024-03-08T12:35:07Z)
• Task-Distributionally Robust Data-Free Meta-Learning [99.56612787882334]
  Data-Free Meta-Learning (DFML) aims to efficiently learn new tasks by leveraging multiple pre-trained models without requiring their original training data. For the first time, we reveal two major challenges hindering their practical deployments: Task-Distribution Shift ( TDS) and Task-Distribution Corruption (TDC)
  arXiv  Detail & Related papers  (2023-11-23T15:46:54Z)
• Amortizing intractable inference in large language models [56.92471123778389]
  We use amortized Bayesian inference to sample from intractable posterior distributions. We empirically demonstrate that this distribution-matching paradigm of LLM fine-tuning can serve as an effective alternative to maximum-likelihood training. As an important application, we interpret chain-of-thought reasoning as a latent variable modeling problem.
  arXiv  Detail & Related papers  (2023-10-06T16:36:08Z)
• Learning Diverse Representations for Fast Adaptation to Distribution Shift [78.83747601814669]
  We present a method for learning multiple models, incorporating an objective that pressures each to learn a distinct way to solve the task. We demonstrate our framework's ability to facilitate rapid adaptation to distribution shift.
  arXiv  Detail & Related papers  (2020-06-12T12:23:50Z)

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.