You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
scripts/setup.sh step 5/7 reports [ok] after the model cache failed #537
Step 5/7 of scripts/setup.sh prints a success label even when the pre-cache failed.
Observed verbatim during an install on 2026-09-09:
=== [5/7] Embedding model ===
Pre-caching sentence-transformers model (one-time ~100MB download)...
[\] running... [|] running... [/] running...Warning: model cache failed (
...
Original error was: No module named 'numpy._core._multiarray_umath'
). Will download on first use.
[ok] Embedding model cached
The model was not cached. The [ok] line contradicts the warning printed immediately above it.
Why it matters
Falling back to a lazy download on first use is a reasonable design, so the step should not be fatal. But the label must say what happened. As written, a reader scanning for [ok] lines concludes the cache is populated when it is not, and the real cause (here a stale numpy extension built for another interpreter) scrolls past unread.
Suggested fix
Distinguish the two outcomes. Report [ok] Embedding model cached only when the pre-cache actually succeeded, and something like [warn] Model not pre-cached, will download on first use otherwise. The step stays non-fatal either way.
Activity
fix(setup): stop labeling a failed model pre-cache [ok] (#537)
fix(setup): stop labeling a failed model pre-cache [ok] (#537) (#543)