fix: Remove feature flag ff_front_dev_1682 to fix Prediction score column showing null#9669
Conversation
…lumn showing null The `annotate_predictions_score` function used `project.model_version` to filter predictions, but this field often stores the ML Backend connection name (title) rather than the actual prediction `model_version`, causing a filter mismatch and null Prediction scores in the Data Manager. The correct logic — filtering by `ml_backends.model_version` — already existed behind feature flag `ff_front_dev_1682_model_version_dropdown_070622_short` (default OFF). This removes the flag and makes the fix unconditional. Fixes HumanSignal#7151
👷 Deploy request for heartex-docs pending review.Visit the deploys page to approve it
|
👷 Deploy request for label-studio-docs-new-theme pending review.Visit the deploys page to approve it
|
✅ Deploy Preview for label-studio-playground ready!
To edit notification comments on pull requests, go to your Netlify project configuration. |
✅ Deploy Preview for label-studio-storybook ready!
To edit notification comments on pull requests, go to your Netlify project configuration. |
matt-bernstein
left a comment
matt-bernstein
left a comment
There was a problem hiding this comment.
Hi! We're investigating the status of this feature flag internally and will get back to you about whether it's safe to merge this change upstream. Glad you were able to find the issue. :)
|
Hi @chenyanchen , the fix for this title-version mismatch has been merged and we expect it to land in label-studio 1.24.0. The fix was actually unrelated to this feature flag, and after investigation, we're staling the feature flag to OFF as it reflects a proposed UI direction for multiple active ML Backends that we did not take. You're welcome to set it ON for your local instance until you pull in the commit that fixes the issue, as long as you're aware it's deprecated. Thanks for raising the issue! |
|
Thanks for the update, @matt-bernstein! Glad to hear the actual fix is already merged. I'll keep the flag ON locally as a workaround until 1.24.0 lands. Appreciate you looking into this! |
3 participants
Summary
ff_front_dev_1682_model_version_dropdown_070622_short(default OFF) and make the corrected prediction score filtering logic unconditionalannotate_predictions_scoreusedproject.model_versionto filter predictions, but this field often stores the ML Backend title (e.g., "My ML Backend") rather than the actualprediction.model_version(e.g., "v1"), causing a filter mismatch →predictions_score: nullin Data Managerml_backends.model_versioninstead — already existed behind the feature flag but was never enabled for open-source usersFiles changed
data_manager/managers.pymodel_versionto filter prediction scorestasks/serializers.pymodel_versionwithproject.model_versionfallbackfeature_flags.jsonflags.tsFF_DEV_1682export (unused in frontend components)tests/conftest.pyFixes #7151
Test plan
test_views_orderingwithpredictions_scoreordering passes (no ML backend, predictions with default emptymodel_version— behavior equivalent)