Uploading FoodExtract-Vision demo app.py

app.py

@@ -86,6 +86,7 @@ Except one model has been fine-tuned on the structured data whereas the other ha
 Notable next steps would be:
 * **Remove the input prompt:** Just train the model to go straight from image -> text (no text prompt on input), this would save on inference tokens.
 * **Fine-tune on more real-world data:** Right now the model is only trained on 1k food images (from Food101) and 500 not food (random internet images), training on real world data would likely significantly improve performance.
+* **Fix the repetitive generation:** The model can sometimes get stuck in a repetitive generation pattern, e.g. "onions", "onions", "onions", etc. We could look into patterns to help reduce this.
 """
 
 demo = gr.Interface(