Repository navigation
Conversation
docs/ci.md pointed at tests/smoketest.sh without saying whether CI runs it. It doesn't: smoke.yaml runs tox -e py3-smoke, which only invokes pytest tests/smoke, never smoketest.sh. The script still covers axes the pytest matrix doesn't (flash-attention, Granite, FSDP+LoRA), so it's orphaned rather than redundant. Notes that explicitly and points at instructlab#705 where the consolidation question is still open. Ref: instructlab#705 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Signed-off-by: Kush Zingade <kush.zingade@gmail.com>
|
Warning Review limit reachedYou've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository. Next included review available in 59 minutes. View limit details
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. |
Maxusmusti
left a comment
There was a problem hiding this comment.
The headline is correct — no workflow invokes tests/smoketest.sh (smoke.yaml and smoke-py312.yaml both go through the run-smoke action → tox -e py3-smoke → pytest tests/smoke). But the added sentence overstates what the script still covers:
- The script is broken, not just orphaned.
test_standard_looppasses--is_granite, which was removed frommain_ds.py's argparse in #257, soparse_args()exits with "unrecognized arguments". The script runs withset -eux -o pipefail, so it aborts there andtest_standard_loop_fsdp_lora(run after it inmain()) is never reached. The script hasn't been touched since 2024-11. - "Granite on/off" isn't a real axis anymore — every test uses
instructlab/granite-7b-lab; the on/off was just that removed flag. - Flash-attention off and FSDP+LoRA genuinely aren't covered by the pytest smoke matrix, so that part is fair — but FSDP+LoRA isn't actually reachable in the script today.
- Smaller points: it names only
smoke.yaml(there's alsosmoke-py312.yaml), and pointing docs at an issue for coverage details will go stale once #705 is closed.
Suggested wording:
There is also a Shell-based smoke test script at
tests/smoketest.sh. It is not run by any CI workflow and is currently stale (e.g. it passes the removed--is_graniteflag); see #705.
The script passes the removed --is_granite flag, so it does not exercise the coverage the previous wording claimed. Signed-off-by: Kush Zingade <kush.zingade@gmail.com>
|
Thanks for the detailed check. I replaced the sentence with your wording: the script is not run by any CI workflow and is currently stale (e.g. it passes the removed |
docs/ci.mdpoints attests/smoketest.shunder "Smoke tests" withoutsaying whether CI runs it. It doesn't:
smoke.yamlinvokestox -e py3-smoke, which (tox.ini) runspytest tests/smoke, nevertests/smoketest.sh. The script still covers axes the pytest matrix doesn't(flash-attention on/off, Granite on/off, FSDP+LoRA), so it's not redundant —
just orphaned and easy to mistake for something CI already checks.
This adds one sentence making that explicit and pointing at #705, where the
consolidation question (fold
smoketest.sh's cases intotest_training_feature_matrix, or delete it) is still open and needs amaintainer decision — actually reconciling the two test surfaces is a bigger
change than a docs PR should carry.
Originally flagged in #705 (comment from 2026-07-29): "There's a second smoke test nobody runs... No workflow calls it anymore, since
smoke.yamlinvokestox -e py3-smokewhich only points attests/smoke/. It's dead as far as CI goes but still checked in and still exercising real coverage."One-line diff,
docs/ci.mdonly.