How to set the num_gpus?

[wmlba]

I am not sure how to set the num_gpus? It's not mentioned anywhere in the docs.

[andremoeller]

Hi @willbadr ,

Thanks for using Amazon SageMaker!

I take it you mean for running MXNet training scripts, right? You don't have to set this since the SageMaker MXNet Container gets this value from the container's environment and injects it into your MXNet script's train() function for you.

Please let us know if this doesn't solve your question. Thanks!

[wmlba]

Thanks a lot for the answer. Yes, that's what I meant. Is there a way to set it manually and pass it to the train() function?

…

On 27 December 2017 at 15:05, andremoeller ***@***.***> wrote: Hi @willbadr <https://github.com/willbadr> , Thanks for using Amazon SageMaker! I take it you mean for running MXNet training scripts, right? You don't have to set this since the SageMaker MXNet Container gets this value from the container's environment and injects it into your MXNet script's train() function for you. Please let us know if this doesn't solve your question. Thanks! — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#34 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ARhcS58Tvcc10OkvRmIXdwwOV8laBzJsks5tEaXbgaJpZM4RNGZA> .

[andremoeller]

Hi @willbadr,

You could use the hyperparameters dictionary in your MXNet script's train() function. Since this is just a dict[string,string], you can choose whatever key you'd like and parse the number of GPUs from that entry's value. You can pass in this dictionary to the MXNet predictor's constructor:

sagemaker-python-sdk/src/sagemaker/mxnet/estimator.py

Lines 20 to 25 in 157d867

	class MXNet(Framework):
	"""Handle end-to-end training and deployment of custom MXNet code."""
	__framework_name__ = "mxnet"
	def __init__(self, entry_point, source_dir=None, hyperparameters=None, py_version='py2', **kwargs):