Open miquelduranfrigola opened 9 months ago
Hi @DhanshreeA
You marked this as Ready, but the issue is not closed. what is the status?
This is not ready and I marked that incorrectly. Currently the --track flag is not incorporated anywhere other than ersilia-self-service.
However, I think we should bundle this with the ersilia inspect
command since that command actually runs the same model a couple of times, for example, with 1, 10, 100 inputs - and an approach like that would be better suited for tracking the computational requirements of an individual model; and collecting the similar data to run a comparison across models.
At present within tracking, we're just taking an average of the memory and CPU percentage consumed regardless of the input sizes - but with that information incorporated, our analyses would be more substantial. So to conclude, we should bundle track with ersilia inspect, and potentially run this within ersilia maintenance
repository.
Hi @DhanshreeA
This is a Batch initiative. Can you detail here the tasks to be completed for this, so I can convert them to issues (maybe in the ersilia-maintenance repository).
Summary
The
ersilia run --track
flag uploads a model digest to an S3 bucket that is eventually monitored by Splunk. It would be good to have acron
job that does a short run to ensure that things are up-and-running. Suggestions are welcome!Objectives
Documentation