Skip to content

Update protos and release 0.1.17#31

Merged
nl917 merged 5 commits intomainfrom
update-sdk
Apr 8, 2026
Merged

Update protos and release 0.1.17#31
nl917 merged 5 commits intomainfrom
update-sdk

Conversation

@nl917
Copy link
Copy Markdown
Contributor

@nl917 nl917 commented Apr 3, 2026

  • Update proto to include newest change from benchmarks

@nl917 nl917 changed the title add new generated files Update protos and release 0.1.17 Apr 3, 2026
@nl917 nl917 requested review from andrewmwang and rosbo and removed request for rosbo April 3, 2026 21:34
@nl917 nl917 requested a review from dolaameng April 3, 2026 21:42
@dolaameng dolaameng requested a review from stevemessick April 3, 2026 22:58
Copy link
Copy Markdown

@dolaameng dolaameng left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM. Thank you @nl917 !

Would like @stevemessick to take a look.

@nl917 nl917 requested a review from erdalsivri April 6, 2026 18:56
@@ -1,4 +1,4 @@
__version__ = "0.1.16"
__version__ = "0.1.17"
Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't know the release process of kaggle-sdk but I suppose there is some manual step here to actually publish this package. I don't know how kaggle-cli imports this package. It used be a manual copy but I am not sure if that has changed. @stevemessick knows better

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Just realized Steve is OOO this week. Requested review from Jonathan instead

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes @nl917 I think these are experimental features only with placeholder code but not real implementation? Checking in the code is to help unblock downstream application. But if this got published to pypi, we probably just wait till everything is implemented by @andrewmwang first.

Copy link
Copy Markdown
Contributor

@jeward414 jeward414 Apr 6, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm also not super familiar with recent implementations/updates to the sdk but I can do a quick look over and see if anything stands out

Copy link
Copy Markdown

@jmasukawa jmasukawa Apr 7, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Once this is merged and a new version of kagglesdk is published to pypi (docs here: https://g3doc.corp.google.com/cloud/kaggle/apis/g3doc/kagglesdk.md#how-to-push-to-pypi), the way Kaggle CLI gets the new package is by bumping it's dependency here: https://github.com/Kaggle/kaggle-cli/blob/main/pyproject.toml#L27

it is currently set to 0.1.16 and we'd change that to 0.1.17.

then we'd need to do a release of Kaggle CLI so our users can pick up the new version. we can test that Kaggle CLI deps pull in the new kagglesdk version locally by:

hatch env remove default
hatch run kaggle --help

(that 2nd line could be any command. we just want to call something to rebuild the CLI)

it'd be good to run the Kaggle CLI tests too to make sure everything is working: https://github.com/kaggle/kaggle-cli?tab=readme-ov-file#tests

btw this assumes in your local CLI env you didn't do anything like pip install kagglesdk --break-system-packages (b/c then it would pull from that global install vs. try to install based on the .toml definition)

@nl917 nl917 requested review from jeward414 and removed request for stevemessick April 6, 2026 19:05
The slug for the task without the user's namespace. (ex: my-task instead of
username/my-task). This should match the slug in the kaggle-benchmarks
decorator in the code.
text (str)
Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@andrewmwang Can you confirm this the text for the .py source code? And on the server side, a .py file with delimiters like # %% will be converted to a ipython notebook properly?

r"""
Attributes:
task_slugs (ApiBenchmarkTaskSlug)
model_slugs (str)
Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@nl917 @andrewmwang what's the behavior here when model_slugs is empty list []. Will it run on all models, or no models?

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

should we make it so that it runs on the default model? that might be the most common option

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

should we make it so that it runs on the default model? that might be the most common option

From a client perspective, the -m here works like a model filter, so if nothing specified, it's probably more intuitive to default to run ALL models. Wdyt?

Copy link
Copy Markdown
Contributor Author

@nl917 nl917 Apr 7, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

My concern is that the model list will only keep increasing. And right now we can only do five runs at a times. will it be too laggy if we run it on all models? I feel like we don't want to make it too easy to do this.

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes that's a good point. In that case can go back to default model.

Copy link
Copy Markdown
Contributor

@stevemessick stevemessick left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

RSLGTM -- I don't usually read the code from kapigen.

Copy link
Copy Markdown

@jmasukawa jmasukawa left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for waiting for my datasets changes. LGTM!

@nl917 nl917 merged commit 6895c97 into main Apr 8, 2026
2 checks passed
@nl917 nl917 deleted the update-sdk branch April 8, 2026 21:37
jmasukawa added a commit to Kaggle/kaggle-cli that referenced this pull request Apr 8, 2026
…ds (#958)

Leverages changes to `DatasetApiService/V1/UpdateDatasetMetadataHandler`
and `DatasetApiService/V1/GetDatasetMetadataHandler` to accept expected
update frequency and user specified sources.

Wait before merging:
- [x] MT changes are deployed:
Kaggle/kaggleazure#41349
- [x] Correct set of kagglesdk changes are made / merged and release is
pushed: Kaggle/kaggle-sdk-python#31

Other changes:
- Pin `virtualenv` version so `./docker-hatch` works
- Run lint / format before making any changes (there was drift)
- Docs updates for datasets / datasets metadata

Local testing:
- [screencast](http://go/scrcast/NTI4NTQxMDQ2ODAwMzg0MHwwNjg0MjViNC1iNQ)

cc: @goeffthomas 

http://b/500108129
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

6 participants