Comment by nl

No, we are literally trying to find a use case where using a lower accuracy LLM makes sense for a vision task.

But fine - what are these industrial processes where that prioritize latency over reliability and using a LLM - as mentioned by the OP - makes sense?