Microsoft AI Watch posted an update
Microsoft’s Azure AI Evaluation Python SDK 1.18.8 fixes three bugs affecting evaluations and conversation handling. The update is dated 8 October in the release notes, published on GitHub the following day.
Why it mattersIt stops legacy safety-annotation polling from retrying terminal HTTP 400 errors until timeout, preserving the service error instead. It also fixes multi-turn evaluations that mishandled per-turn token counts, and conversation formatting that failed when system messages used typed text blocks. For teams using the SDK, these are practical reliability fixes, not a new feature tour. If your Azure evaluation runs have been tripping over any of those cases, this is the patch to look at.
Discuss: Which fix is likeliest to save your team the most time: clearer errors, correct token reporting, or better conversation handling?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.
Your turn
Pull up a chair.
Write first. We’ll sort the introductions when you submit.