Open pacospace opened 3 years ago
/assign @pacospace
More details: https://chat.google.com/room/AAAAVjnVXFk/YwV3qk8eoxA
Ideas:
pre-requisities
advised_manifest_changes
https://thoth-station.ninja/docs/developers/adviser/prescription/wraps.html#wrap-run-advised-manifest-changesextra steps (not required) to be converted to issues
pre-requisities
- [ ] create prescription as suggested by @fridex using
advised_manifest_changes
https://thoth-station.ninja/docs/developers/adviser/prescription/wraps.html#wrap-run-advised-manifest-changes- [ ] extend Thamos with a new flag to get the base image recommended, only show to user initially
extra steps (not required)
to be converted to issues
- [ ] create spike on automation of creation of the prescriptions for ps-stacks and main libraries involved (e.g PyTorch and open-cv for ps-cv)
- [ ] next extension for Thamos to actually apply the base image recommended
Thanks for the meeting minutes write-up. First, we can start with prescriptions - they will be easy to write, maintain and can state the main packages for which we want to recommend the base image (e.g. pytorch
stack).
Besides that, we should prioritize fix https://github.com/thoth-station/thoth-application/issues/1788 as it is causing wrong data to be synced and associated with the container images we have in the database. It will unlock https://github.com/thoth-station/thoth-application/issues/1788 and a possibility to query packages in container images we build in a separate pipeline unit that can recommend a base image with an exact/partial match for the stack. The prescriptions scenario is still applicable as this solution just extends what we offer to users.
On the user's side, we can have the mentioned flag that will apply changes to the .thoth.yaml
and eventually .aicoe-ci.yaml file. We should revisit the current design of advised_manifest_changes
to distinguish what data are propagated from the recommender, how they are recognized and subsequently applied to the corresponding files.
pre-requisities
- [ ] create prescription as suggested by @fridex using
advised_manifest_changes
https://thoth-station.ninja/docs/developers/adviser/prescription/wraps.html#wrap-run-advised-manifest-changes- [ ] extend Thamos with a new flag to get the base image recommended, only show to user initially
extra steps (not required) to be converted to issues
- [ ] create spike on automation of creation of the prescriptions for ps-stacks and main libraries involved (e.g PyTorch and open-cv for ps-cv)
- [ ] next extension for Thamos to actually apply the base image recommended
Thanks for the meeting minutes write-up. First, we can start with prescriptions - they will be easy to write, maintain and can state the main packages for which we want to recommend the base image (e.g.
pytorch
stack).
Thanks @fridex, let's do it!!
Besides that, we should prioritize fix thoth-station/thoth-application#1788 as it is causing wrong data to be synced and associated with the container images we have in the database.
Sounds good, @harshad16 what is ETA for the mentioned issue?
It will unlock thoth-station/thoth-application#1788 and a possibility to query packages in container images we build in a separate pipeline unit that can recommend a base image with an exact/partial match for the stack. The prescriptions scenario is still applicable as this solution just extends what we offer to users.
@fridex it will unlock thoth-station/thoth-application#1788 ?
Do we still want to use prescriptions in that case?
On the user's side, we can have the mentioned flag that will apply changes to the
.thoth.yaml
and eventually .aicoe-ci.yaml file. We should revisit the current design ofadvised_manifest_changes
to distinguish what data are propagated from the recommender, how they are recognized and subsequently applied to the corresponding files.
Perfect!
@fridex it will unlock thoth-station/thoth-application#1788 ?
thoth-station/thoth-application#1788 will unlock https://github.com/thoth-station/adviser/issues/1533 (also discussed in https://github.com/thoth-station/adviser/issues/1943).
Do we still want to use prescriptions in that case?
Yes, they provide a different view of container images and Python packages shipped in them.
@fridex it will unlock thoth-station/thoth-application#1788 ?
thoth-station/thoth-application#1788 will unlock thoth-station/adviser#1533 (also discussed in thoth-station/adviser#1943).
Thanks, in the above comment was not linked, now I get it :)
Do we still want to use prescriptions in that case?
Yes, they provide a different view of container images and Python packages shipped in them.
Perfect, then we also want to introduce a mechanism to automatically create prescriptions from ps-stacks analysis.
Perfect, then we also want to introduce a mechanism to automatically create prescriptions from ps-stacks analysis.
A simple script that takes direct dependencies from Pipfiles and generates corresponding prescriptions out of them (as direct dependencies are the ones that in most cases define the container image use) should be good enough here. Later it could be plugged into a bigger workflow if necessary.
pre-requisities
- [ ] create prescription as suggested by @fridex using
advised_manifest_changes
https://thoth-station.ninja/docs/developers/adviser/prescription/wraps.html#wrap-run-advised-manifest-changes- [ ] extend Thamos with a new flag to get the base image recommended, only show to user initially
extra steps (not required) to be converted to issues
- [ ] create spike on automation of creation of the prescriptions for ps-stacks and main libraries involved (e.g PyTorch and open-cv for ps-cv)
- [ ] next extension for Thamos to actually apply the base image recommended
Issues created based on conversations with @fridex .
blocked by https://github.com/thoth-station/python/issues/418 and quay issues.
/kind documentation /priority critical-urgent
Missing steps from Thoth side
Missing steps from Operate First side:
Tutorial specific:
/assign @mayaCostantini
Issues go stale after 90d of inactivity.
Mark the issue as fresh with /remove-lifecycle stale
.
Stale issues rot after an additional 30d of inactivity and eventually close.
If this issue is safe to close now please do so with /close
.
/lifecycle stale
Stale issues rot after 30d of inactivity.
Mark the issue as fresh with /remove-lifecycle rotten
.
Rotten issues close after an additional 30d of inactivity.
If this issue is safe to close now please do so with /close
.
/lifecycle rotten
Is your feature request related to a problem? Please describe. As Data Scientist,
I have a notebook for Computer Vision that I want to use in ODH.
As Data Scientist,
I have a stack with opencv and pytorch. I want to know which image should I use to run my Computer Vision Application.
High-level Goals Have a tutorial on using a recommended base image for my ML application that can be run on Operate First with Meteor.
Describe the solution you'd like Create steps in the tutorial.
Pre-requisite:
Steps:
Describe alternatives you've considered
Additional context Related-To: https://github.com/thoth-station/core/issues/308
Acceptance Criteria