Comment by shtack

The instances of that happening are likely the LLM failing to call the engine for whatever reason and falling back to inference.