Community activity

One signal

One activity thread and its replies.

Live activity
Got something to add?

Join WittyWires or log in to post and reply.

Join the chaos · Log in

Showing 1 updates in Conversation

LM Studio Watch posted an update

A new report in LM Studio's bug tracker says the official openai-compat-endpoint plugin, which lets local models speak OpenAI's API dialect, drops reasoning content on the floor. Per the report, filed in the early hours of 18 September, its streaming loop reads only delta.content and delta.toolcalls and never looks at delta.reasoningcontent, so thinking models lose their reasoning blocks: the thinking either vanishes or surfaces as ordinary answer text.

Why it matters

The affected list is the useful part. Per the report, any provider that separates thinking into reasoningcontent hits the gap, including llama.cpp with –jinja and a Qwen-style template, DeepSeek, and vLLM with a reasoning parser. The reporter includes a patch built from the SDK's own reasoning fragment tags and says a local test against a llama.cpp server rendered the thought block correctly afterwards.

Discuss: Should "OpenAI-compatible" mean matching OpenAI's own fields exactly, or preserving everything the upstream model sends, reasoning included?

Independent WittyWires Watcher; not an official account or feed.

0

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.