how to increase the storage of host instance

Question

The parameters of a big neural network model can be huge. But the largest storage size of a host instance is only 30G, according to https://docs.aws.amazon.com/sagemaker/latest/dg/host-instance-storage.html. Is there a way to increase the storage volume? I have a model (embeddings) that is very close to 30G and caused a no space error when deploying.   
  
Thanks!

Accepted Answer

The disk size is currently not configurable for SageMaker Endpoints with EBS backed volumes. As a workaround, please use instances with ephemeral storage for your SageMaker endpoint.  
  
Example instance types with ephemeral storage:  
- m5d instances: https://aws.amazon.com/ec2/instance-types/m5/  
- c5d instances: https://aws.amazon.com/ec2/instance-types/c5/  
- r5d instances: https://aws.amazon.com/ec2/instance-types/r5/  
  
The full list of Amazon SageMaker instance types can be accessed here: https://aws.amazon.com/sagemaker/pricing/instance-types/

Answer

Thanks! Using a x5d instance solves the issue.  
  
And a quick note: even though I could download the big model to endpoint now, I got timeout error when loading the model in the endpoint. After some trial and error, I solved it by increasing the timeout value and reducing the number of worker by setting the environment variables. To do it, pass this dict   
  
env={"SAGEMAKER_MODEL_SERVER_WORKERS":"1",  
         "SAGEMAKER_MODEL_SERVER_TIMEOUT":"1800"}  
  
when creating the model object.

how to increase the storage of host instance

Conteúdo relevante