Pytorch.fit() not working

[LightToYang]

Describe the bug

inputs are dirs to s3

from sagemaker.pytorch import PyTorch

estimator = PyTorch(entry_point='mnist.py',
                    role=role,
                    py_version='py3',
                    framework_version='1.8.0',
                    instance_count=2,
                    instance_type='ml.c5.2xlarge',
                    hyperparameters={
                        'epochs': 1,
                        'backend': 'gloo'
                    })
estimator.fit({'train': inputs, 'test': inputs})

dir still

"channel_input_dirs": {
        "test": "/opt/ml/input/data/test",
        "train": "/opt/ml/input/data/train"
    },

but not
s3:// .....

And while directly set dataset dir to inputs s3:// ....., dataset still not found with torchvision.datasets.MNIST

System information
A description of your system. Please provide:

• sagemaker version: 2.59.3.post0:
• PyTorch, MNIST tutorial:
• Python 3 (PyTorch 1.6 Python 3.6 GPU Optimized)

[mufaddal-rohawala]

Apologies for a late follow up on this issue. It seems like you are facing issues configuring S3 Inputs for a Pytorch Estimator. You can refer to following example for a helper to configure S3 input channels for Pytorch Estimator. Closing this issue for now, please feel free to reopen if you continue to see this issue with the latest sagemaker version.