# Merge then Realign: Simple and Effective Modality-Incremental Continual Learning for Multimodal LLMs

arXiv: https://arxiv.org/abs/2503.07663 · HTML: https://mlsanity.com/paper/2503.07663/

## Key facts

- Cited by 6 papers in the map (data through 14 Oct 2026).
- Placed in Prompt learning and adaptation of CLIP, one of 16 directions in Multimodal LLMs / vision-language models.
- Builds on 48 of its 70 references that are in the map.
- First posted on arXiv 8 Mar 2025.

## Placed in

- Multimodal LLMs / vision-language models / [Prompt learning and adaptation of CLIP](https://mlsanity.com/mllm/prompt-learning-and-adaptation-of-clip/)

## Builds on (most cited first, 30 of 48)

- [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://mlsanity.com/paper/1810.04805/) (2018)
- [Microsoft COCO: Common Objects in Context](https://mlsanity.com/paper/1405.0312/) (2014)
- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Overcoming catastrophic forgetting in neural networks](https://mlsanity.com/paper/1612.00796/) (2016)
- [Improved Baselines with Visual Instruction Tuning](https://mlsanity.com/paper/2310.03744/) (2023)
- [Flamingo: a Visual Language Model for Few-Shot Learning](https://mlsanity.com/paper/2204.14198/) (2022)
- [InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning](https://mlsanity.com/paper/2305.06500/) (2023)
- [CIDEr: Consensus-based Image Description Evaluation](https://mlsanity.com/paper/1411.5726/) (2014)
- [Video-LLaVA: Learning United Visual Representation by Alignment Before Projection](https://mlsanity.com/paper/2311.10122/) (2023)
- [OK-VQA: A Visual Question Answering Benchmark Requiring External Knowledge](https://mlsanity.com/paper/1906.00067/) (2019)
- [Experience Replay for Continual Learning](https://mlsanity.com/paper/1811.11682/) (2018)
- [TIES-Merging: Resolving Interference When Merging Models](https://mlsanity.com/paper/2306.01708/) (2023)
- [Robust fine-tuning of zero-shot models](https://mlsanity.com/paper/2109.01903/) (2021)
- [An Empirical Investigation of Catastrophic Forgetting in Gradient-Based Neural Networks](https://mlsanity.com/paper/1312.6211/) (2013)
- [Language Models are Super Mario: Absorbing Abilities from Homologous Models as a Free Lunch](https://mlsanity.com/paper/2311.03099/) (2023)
- [VILA: On Pre-training for Visual Language Models](https://mlsanity.com/paper/2312.07533/) (2023)
- [Three scenarios for continual learning](https://mlsanity.com/paper/1904.07734/) (2019)
- [Progress & Compress: A scalable framework for continual learning](https://mlsanity.com/paper/1805.06370/) (2018)
- [BEATs: Audio Pre-Training with Acoustic Tokenizers](https://mlsanity.com/paper/2212.09058/) (2022)
- [What is being transferred in transfer learning?](https://mlsanity.com/paper/2008.11687/) (2020)
- [PointLLM: Empowering Large Language Models to Understand Point Clouds](https://mlsanity.com/paper/2308.16911/) (2023)
- [Orthogonal Subspace Learning for Language Model Continual Learning](https://mlsanity.com/paper/2310.14152/) (2023)
- [New Insights on Reducing Abrupt Representation Change in Online Continual Learning](https://mlsanity.com/paper/2104.05025/) (2021)
- [AnyGPT: Unified Multimodal LLM with Discrete Sequence Modeling](https://mlsanity.com/paper/2402.12226/) (2024)
- [Boosting Continual Learning of Vision-Language Models via Mixture-of-Experts Adapters](https://mlsanity.com/paper/2403.11549/) (2024)
- [OneLLM: One Framework to Align All Modalities with Language](https://mlsanity.com/paper/2312.03700/) (2023)
- [VITA: Towards Open-Source Interactive Omni Multimodal LLM](https://mlsanity.com/paper/2408.05211/) (2024)
- [Fine-tuned Language Models are Continual Learners](https://mlsanity.com/paper/2205.12393/) (2022)
- [Pushing Mixture of Experts to the Limit: Extremely Parameter Efficient MoE for Instruction Tuning](https://mlsanity.com/paper/2309.05444/) (2023)
- [X-LLM: Bootstrapping Advanced Large Language Models by Treating Multi-Modalities as Foreign Languages](https://mlsanity.com/paper/2305.04160/) (2023)

## Built on by (most cited first, 6 of 6)

- [FOREVER: Forgetting Curve-Inspired Memory Replay for Language Model Continual Learning](https://mlsanity.com/paper/2601.03938/) (2026)
- [Adaptive and Resource-efficient Agentic AI Systems for Mobile and Embedded Devices: A Survey](https://mlsanity.com/paper/2510.00078/) (2025)
- [MACS: Modality-Aware Capacity Scaling for Efficient Multimodal MoE Inference](https://mlsanity.com/paper/2605.05225/) (2026)
- [Fast-Slow Efficient Training for Multimodal Large Language Models via Visual Token Pruning](https://mlsanity.com/paper/2602.03815/) (2026)
- [Continual Vision-Language Learning for Remote Sensing: Benchmarking and Analysis](https://mlsanity.com/paper/2604.00820/) (2026)
- [VFA: Empowering Multilingual MLLMs via Vision-Free Adaptation](https://mlsanity.com/paper/2608.26155/) (2026)

## Abstract (arXiv)

Recent advances in Multimodal Large Language Models (MLLMs) have enhanced their versatility as they integrate a growing number of modalities. Considering the heavy cost of training MLLMs, it is efficient to reuse the existing ones and extend them to more modalities through Modality-incremental Continual Learning (MCL). The exploration of MCL is in its early stages. In this work, we dive into the causes of performance degradation in MCL. We uncover that it suffers not only from forgetting as in traditional continual learning, but also from misalignment between the modality-agnostic and modality-specific components. To this end, we propose an elegantly simple MCL paradigm called "MErge then ReAlign" (MERA) to address both forgetting and misalignment. MERA avoids introducing heavy model budgets or modifying model architectures, hence is easy to deploy and highly reusable in the MLLM community. Extensive experiments demonstrate the impressive performance of MERA, holding an average of 99.84% Backward Relative Gain when extending to four modalities, achieving nearly lossless MCL performance. Our findings underscore the misalignment issue in MCL. More broadly, our work showcases how to adjust different components of MLLMs during continual learning.