#dual-learner approach

WINGS introduces a dual-learner architecture that balances attention between text and images in multimodal large language models, significantly reducing text-only forgetting and boosting performance across benchmarks.