Distributed training & inference for multimodal perception: shared embeddings across text, images, video, audio, code & graphs from 64 to 1024 dimensions 🖼️