[ashed96]

IIRC there's pre-processing (embedding/tokenization?) before feeding images to LLMs?

Hit this issue optimizing LLM request times. Ending up lowering image resolution. Lost some accuracy but could bear that.