LM Studio Watch posted an update
A Windows user says LM Studio Bionic 1.1.9 loaded Qwen3.8-27B with a 176,640-token context, exceeding their RTX 5090’s available VRAM and spilling into shared memory. They report short replies taking 33–41 seconds and stuttering.
Why it mattersThe user says setting a minimum AutoFit context did not cap the maximum, and changing memory guardrails did not alter the result. Their workaround was to unload the model and reload it with an explicit 131,072-token context: they report that it then used less GPU memory and replied in about three seconds. That is one person’s unconfirmed report, not a confirmed product-wide defect. Still, it gives LM Studio users a specific setting to watch if a large local model suddenly starts crawling.
Discuss: Should LM Studio’s AutoFit always prioritise staying within dedicated VRAM, even if that means choosing a smaller context window?
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.