How is Azure ML custom Environment Autosave version generated

Question

Hi,
The training notebook points to the specified environment version but once the run is submitted, the run picks up the env with version=autosave... instead of the version I had specified which leads to failure of the run submitted.
Could you help me understand how this autosave environment version gets created once the run is submitted.

Answer 1

@Rajesh Kumar S Thanks for the question. Can you please add more details about the code that you are trying.
Here is the link to setup runconfig environment configuration we have to do is for configuring the Python runtime environment which is defined by a separate yaml file (conda_dependencies.yml).
https://github.com/microsoft/MLOps/blob/master/examples/cli-train-deploy/generate-runconfig.py

Answer 2

Hi, I am having the same question. Those are from web UI ->Environments-> SELECT A SAVED ENVIRONMENT -> Version dropdown. See the screenshot.

Answer 3

I am facing the same issue. @Ramr-msft do you have any comments on this? @Rajesh Kumar S Did you figure out a solution?

Answer 4

Hello,

I have done some tests
For 3 different pip requirements files (A.txt, B.txt and C.txt) with different pip dependencies, I have constructed associated envA, envB, envC with the Environment class from AzureML.

I have register and build envA from SDK then submitted a job
I have register EnvB from SDK then submitted a job
I have submitted a job with EnvC not registered and not built.

Key Results :
Autosave is created when you submit a run without having registered your environment

Furthermore, on others Experiment from my work, I have noticed that having autosaved version while calling a specific one might be due to the fact that they are pointing to the same built image