We need some utils function to be able to load/save models.
- Saving a model: saving a physical model is to store the config and physical parameters, whereas for ml models, the common save procedure in torch can apply i.e.
model.state_dict() and model metadata.
- Loading a model: Engine should be able to load a model (physical or ml)
- Format: torch.save/load is using pickle internally. and onnx has a complex setup. A better option is to use SafeTensors.
We need some utils function to be able to load/save models.
model.state_dict()and model metadata.