Skip to content

Improve dataset preload startup UX - #2457

Open
hannahwestra25 wants to merge 2 commits into
microsoft:mainfrom
hannahwestra25:hannahwestra25-fix-dataset-preload-startup
Open

Improve dataset preload startup UX#2457
hannahwestra25 wants to merge 2 commits into
microsoft:mainfrom
hannahwestra25:hannahwestra25-fix-dataset-preload-startup

Conversation

@hannahwestra25

@hannahwestra25 hannahwestra25 commented Aug 21, 2026

Copy link
Copy Markdown
Contributor

What changed

  • Default config now fetches only the datasets a scenario needs.
  • Full dataset preload is still available as an opt-in for cache warming or offline use.
  • Startup timeout errors now give a helpful hint when dataset preloading is still running.
  • Added focused regression tests.

LoadDefaultDatasets behavior and the default timeout are unchanged.

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
@hannahwestra25
hannahwestra25 marked this pull request as ready for review August 21, 2026 19:02
- name: scorer
- name: technique
- name: load_default_datasets
# Optional full preload/cache warming; scenarios fetch requested datasets on demand.

@jsong468 Justin Song (jsong468) Aug 21, 2026

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.

Suggested change
# Optional full preload/cache warming; scenarios fetch requested datasets on demand.
# Optional full preload/cache warming; scenarios fetch requested datasets on demand and add them to memory.

preload_message = ""
if "loading datasets" in log_tail.casefold():
preload_message = (
" Dataset preloading is still in progress. Remove load_default_datasets from the config "

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.

cp: Finding “loading datasets” in the recent log proves that preload occurred, but not necessarily that it is still running—the completed tqdm line can remain in the log while startup later hangs elsewhere. Consider wording this as “Dataset preloading was detected in the recent logs” or “may still be in progress” to avoid a misleading diagnosis.

@@ -8,4 +8,5 @@ initializers:
- scorer
- name: scorer
- name: technique

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.

fine to remove, but we may want to modify the datasets the garak hallucination scenario uses by default so we can make the default run go faster. Fine to do in a followup

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.

3 participants