Is it possible to perform local dev on a CPU-only machine on HF/sagemaker?

I’m trying to dev locally on sagemaker.huggingface.HuggingFace before moving to sagemaker for actual training. I set up a

HF_estimator = HuggingFace(entry_point='train.py', instance_type='local' ...)

And called HF_estimator.fit()

In train.py im simply printing and exiting to see if it will work. However I ran into this:

ValueError: Unsupported processor: cpu. You may need to upgrade your SDK version (pip install -U sagemaker) for newer processors. Supported processor(s): gpu.

Is it possible to bypass this for local development?

Hello @plamb,

Currently, there are only GPU DLCs for Training Hugging Face Models. What you could is execute your train.py with python3 train.py and pass in your hyperparameters (dict) as cli args with --key value