Loading Models from Cloud

aitoolbox.cloud.AWS.model_load and aitoolbox.cloud.GoogleCloud.model_load modules provide logic to conveniently download the previously saved model checkpoint from the cloud storage and initialize local model weight. This in effect automatically jump-starts the local model from the saved checkpoint and makes it ready for inference use or further training.

Currently available cloud model loading is for the PyTorch and supports AWS S3 and Google Cloud Storage:

• aitoolbox.cloud.AWS.model_load.PyTorchS3ModelLoader

• aitoolbox.cloud.GoogleCloud.model_load.PyTorchGoogleStorageModelLoader

To load the model from cloud, first, the user needs to initialize the model object and then give it to the cloud model loader. Model loader will download the saved model from the cloud and use its weights to initialize the provided local model. Furthermore the model loader can also initialize the optimizer and the AMP in case the local model is not going to be used just for inference but for further model training.