StrLoRA: Towards Streaming Continual Visual Instruction Tuning for MLLMs

TL;DR

StrLoRA introduces a novel expert routing framework for streaming continual visual instruction tuning, enabling multimodal models to learn from dynamic, interleaved data streams while mitigating forgetting.

Contribution

The paper proposes StrLoRA, a two-stage expert routing method with regularization, to improve continual learning in a realistic streaming setting for multimodal models.

Findings

StrLoRA outperforms existing methods on the StrCVIT benchmark.

It effectively distinguishes and adapts to heterogeneous task samples.

The approach enhances model capabilities in evolving data streams.

Abstract

Continual Visual Instruction Tuning (CVIT) enables Multimodal Large Language Models to incrementally acquire new abilities. However, existing CVIT methods operate under a restrictive task-incremental setting, where each training phase corresponds to a single, predefined task. This does not reflect real-world conditions, where data arrives as a continuous stream of interleaved and dynamically evolving tasks. To bridge this gap, we introduce Streaming CVIT (StrCVIT), a more general and realistic setting where models learn from a stream of data chunks containing a dynamic mixture of tasks. In StrCVIT, a model must simultaneously acquire new abilities, reinforce recurring abilities, and mitigate forgetting. Existing CVIT methods fail here as they cannot reliably distinguish or adapt to the heterogeneous task samples within each chunk. We therefore propose StrLoRA, a regularized two-stage expert routing framework. StrLoRA first performs task-aware expert selection using the textual instruction to activate a sparse subset of relevant experts, reducing cross-task interference. It then applies token-wise expert weighting within this subset, where contribution weights are computed via cross-modal attention between local visual tokens and the global instruction representation. To maintain stability across the non-stationary stream, a routing-stability regularization aligns current routing distributions with a historical exponential moving average reference. Extensive experiments on a newly developed StrCVIT benchmark show that StrLoRA substantially outperforms existing methods, effectively enhancing model's abilities from continuously evolving data streams. The code is available at https://github.com/chanceche/StrCVIT.