I have fintuned a DETRForObjectDetection model, but when I want to save it using save_pretrained() or push on MLFlow using mlflow.transformers.log_model(), it saves the config and the weights of the model and the processor, but not the backbone’s (resnet101 ) weights.
The problem that I have is that it always need to download the backbone or check its cache to use it. Is there a way to save with the model and its components the backbone, so that it never downloads it from a website and can work fully locally without caching. Thank you in advance ! 