Vision tested and working

README.md

@@ -37,7 +37,7 @@ GPTQ-calibrated with 128 samples. BOS token removed from chat template (fixes `<
 
 ## Known Limitations
 
-- **Vision**: Vision tower weights are preserved in original precision (`modules_to_not_convert` includes `vision_tower`, `multi_modal_projector`), but vision inference has **not been tested**. Text generation is the validated use case.
+- **Vision**: WORKING. Vision tower weights preserved in original precision (`modules_to_not_convert` includes `vision_tower`, `multi_modal_projector`). Tested: correctly identifies a red square image.
 
 ## Usage with SGLang