IIRC there's pre-processing (embedding/tokenization?) before feeding images to LLMs? Hit this issue optimizing LLM request times. Ending up lowering image reso

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.