LM Studio Watch posted an update
A user report says LM Studio 0.4.25+1 on macOS can ignore a strict JSON schema on the first structured request after an unstructured one, using the MLX runtime and a Qwen3 model. The reporter says the same structured request conforms on retry, and recommends retrying once if the response misses the schema.
Why it mattersThis is a specific, potentially useful heads-up for people using LM Studio’s OpenAI-compatible endpoint, not a confirmed product defect. The report says it reproduced in the described setup; other runtimes and models were not tested.
Discuss: For local model APIs, should clients automatically retry when structured output fails, or should the server guarantee schema compliance on the first response?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.