That seems to be just for LLMs, not visual. I'm wanting to go from images of maths notation (photos, scans, digital handwriting) to formulas in Latex or MathML or something. Qwen2-VL can do it, but it's pretty heavyweight for just that.