StrLoRA: Towards Streaming Continual Visual Instruction Tuning for MLLMs

In three linesStrLoRA introduces a streaming continual visual instruction tuning framework for MLLMs. Unlike existing methods restricted to predefined tasks, StrCVIT handles data streams with dynamic, interleaved tasks. StrLoRA employs two-stage expert routing with task-aware selection and token-wise weighting, stabilized via routing-stability regularization.

Summary generated by Claude — human-verified