tfx icon indicating copy to clipboard operation
tfx copied to clipboard

Are we able to override version name?

Open snehpandya123 opened this issue 4 years ago • 10 comments

https://github.com/tensorflow/tfx/blob/9b569cd96b38a15a7751db236c80fa2cf66bcb6a/tfx/extensions/google_cloud_ai_platform/pusher/executor.py#L97

snehpandya123 avatar Aug 20 '20 12:08 snehpandya123

@snehpandya123, Can you please elaborate on what exactly you want to do. Thanks!

rmothukuru avatar Aug 21 '20 06:08 rmothukuru

@snehpandya123, Can you please respond to the above comment. Thanks!

rmothukuru avatar Aug 28 '20 09:08 rmothukuru

Can we specify name of model version instead of v15**** which is default.

snehpandya123 avatar Sep 01 '20 15:09 snehpandya123

Can you explain the use case of a different version string? The unspoken requirements we had on this version string are:

  • unique
  • monotonic increasing
  • easy to parse.

Timestamp makes pretty good sense. If you have a different use case, can you explain that info is, and how you plan to pass it through the pipeline DSL?

zhitaoli avatar Sep 02 '20 16:09 zhitaoli

In our case we have a blue/green deployment and have 2 models, Model1(east region) and Model2(central region) and both have 2 versions Version1 and Version2. At given time Version1 is Green and next week we will have retrained model deployed to Version2 ( Delete and Recreate ) and after deployment Version1 becomes passive.

snehpandya123 avatar Sep 02 '20 18:09 snehpandya123

Thanks for the feedback. Adding owner of pusher @chongkong to the discussion, specially since he has a pending TODO on "introducing version".

zhitaoli avatar Sep 03 '20 16:09 zhitaoli

Thank You.

snehpandya123 avatar Sep 03 '20 17:09 snehpandya123

Sorry for the late followup. At the moment we have only automatic versioning scheme (and only with timestamp). If your demand is urgent our recommendation is to create your own custom component that consumes model to publish it with your custom version name. Feel free to consult us for the best practice for getting the stateful version value.

chongkong avatar Sep 13 '20 14:09 chongkong

If we need to push the model to multiple regions, every time it runs the Pusher, it creates a new timestamp as the model_version, which cause the same model has different versions in different regions. Is there any suggestions on that?

zhongdai avatar Sep 04 '22 10:09 zhongdai

We haven't developed alternative Versioning scheme yet, but we can introduce another one, depending on your purpose and needs. For example we can have external versioning scheme that accepts string version number from another custom components (that you create) to be used. Or versioning based on the pipeline_run which would be consistent across the single run. You can ask for a feature request, but unfortunately this would be low priority for the team 🥲.. The fastest option for you would be implementing your own custom pusher (maybe you can inherit from the original pusher implementation) with the functionality you want.

chongkong avatar Sep 22 '22 09:09 chongkong

This issue has been marked stale because it has no recent activity since 7 days. It will be closed if no further activity occurs. Thank you.

github-actions[bot] avatar May 03 '23 01:05 github-actions[bot]

This issue was closed due to lack of activity after being marked stale for past 7 days.

github-actions[bot] avatar May 11 '23 01:05 github-actions[bot]