#language and vision models

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.