Multimodal Neurons in Pretrained Text-Only Transformers

Open in new window