deploy endpoint in complete local mode #2182

Activity

1. ChoiByungWook

   added

   component: local mode

   type: documentation

   type: question

   on Mar 2, 2021
2. ChoiByungWook commented on Mar 3, 2021

   

   ChoiByungWook

   on Mar 3, 2021

   Last edited by ChoiByungWook

   Contributor

   More actions

   Hey @anotinelg,

   Thanks for bringing this issue to our attention.

   This looks like a gap in our Python SDK for supporting true local mode when using our Framework Estimator/Model.

   There doesn't seem to be a way to disable repack_model through the Framework Estimator/Model constructors.

   I think one workaround that may work is to use the Estimator/Model class and point the image_uri to the corresponding image that is used in the Framework Estimator you were attempting to run local mode with, as the base Model class shouldn't call repack_model.

   sagemaker-python-sdk/src/sagemaker/model.py

   Line 282 in c4d71f5

   return sagemaker.container_def(self.image_uri, self.model_data, self.env)

   I will note this in backlog, which is prioritized based on customer feedback.
3. anotinelg commented on Mar 4, 2021

   

   anotinelg

   on Mar 4, 2021

   Author

   More actions

   Actually i am using the MxnetModel which will use the Framework Model and hence the method prepare_container_def which calls the upload_code. I am not sure i understand what you propose. Instead of using MXNetModel, i should directly use the Model class?
4. ChoiByungWook commented on Mar 4, 2021

   

   ChoiByungWook

   on Mar 4, 2021

   Last edited by ChoiByungWook

   Contributor

   More actions

   Actually i am using the MxnetModel which will use the Framework Model and hence the method prepare_container_def which calls the upload_code. I am not sure i understand what you propose. Instead of using MXNetModel, i should directly use the Model class?

   Yeah, the MXNetModel class ends up extending the FrameworkModel class, which then extends the Model class.

   I believe the base Model class doesn't call repack_model, however you would lose out on a few things that MXNetModel class does for you such as the image_uri, repacking of the model with your inference script. These things you would have to supply to the base Model class.
5. bveeramani

   closed this as completedon May 19, 2021
6. aws

   locked and limited conversation to collaborators on May 19, 2021