How do you use the custom generator to train the TensorFlow model on PageMaker?

I am trying to train a TensorFlow custom classification model using a custom data generator on SageMaker Studio. However, the notebook did not utilize the GPU to train the model.

My instance type:

I have stored the images in a zip file on an S3 bucket and unzipped them during training, using the temporary path in the custom generator for training.

I want to add data augmentation while training the model.

The example below loads the images in numpy format and passes them directly to the training model. I want to load the images from the S3 bucket in zip format and train with a custom generator and augmentation.

[Link to the example](https://sagemaker-examples.readthedocs.io/en/latest/sagemaker-experiments/sagemaker_job_tracking/tensorflow_script_mode_training_job.html)

Start with the linked TensorFlow script-mode training example and compare its data-loading flow with the requested S3 zip, custom generator, and augmentation setup. Done would be a clear example or documented guidance showing how this setup trains the model on SageMaker Studio and uses the GPU.

Written by the indexing model from the issue text.