Add historical data visualization features

#7
by badaoui HF Staff - opened

I made some changes to improve the CI dashboard, building on Manuel's work on adding historical data graphs. Here’s the link: https://huggingface.co/spaces/badaoui/Tcid

What I added / improved:

  • First Seen Date of failing test : in order to provide immediate context on which PRs might have caused the failure.
  • “Failing Models” toggle : Shows only failing models. You can also filter by hardware (AMD or NVIDIA) to display failing models for a specific platform.
  • New Regression detection : Compares the set of failed tests from the latest run against the previous run. Any test failing today but not yesterday is flagged as a regression and needs immediate attention.
  • Fixed historical failure rates graph : Now matches the main dashboard calculation by including ERROR tests in failures and excluding SKIPPED tests from the total. Formula is now (failed + error) / (passed + failed + error) instead of failed / (passed + failed + skipped).
  • Navigation fixes : Resolved small issues when browsing pages (e.g., clicking on a model → history view → Summary → history view now correctly shows the summary page).
  • Improved visualizations : Switched to Plotly for better graphs.
  • Added logos
badaoui changed pull request status to open
Transformers Community org

Awesome, thanks for the contribution! I have a few comments, I will organize according to the bullet points:

  • First Seen Date of failing test nothing: to say here, good idea, good UI, neat
  • New Regression detection: very cool, the location of the button is a bit awkward but then again I have no other alternative as of now. If you have one at some point aybe open a PR.
  • Fixed historical failure rates graph : cool thank you
  • Navigation fixes : nice!
  • Improved visualizations : for historical view only right?
  • Added logos: this part looks a bit tacky to me, could you revert please? Thanks!

Other than that, maybe you can re-use some code that you added in different places? The diff is quite large, that's why I am asking. Thanks!

Thanks for your review, Rémi.

  • Yes, once I find a better placement for the new regression button, I’ll open a PR for it. I was thinking about placing it at the top of the page, but I don’t want it to draw too much attention away from the main CI dashboard.
  • Yes, I used Plotly only for the historical graphs.
  • Alright, I’ll revert the logo additions as requested and check for code reuse opportunities to reduce duplication.
Transformers Community org

Talked offline, this is in a good place to merge, and it looks good on the fork. Merging, thank you!

ror changed pull request status to merged

Sign up or log in to comment