An ONNX-based implementation of the CLIP model that doesn't depend on torch or torchvision. - View it on GitHub