You don’t need a language model for that, and arguably shouldn’t use one if there’s a possibility of the model hallucinating the wrong answer. And the model is

You don’t need a language model for that, and arguably shouldn’t use one if there’s a possibility of the model hallucinating the wrong answer. And the model is only as good as the information it is given, so if there is allergen info available, why not just display it directly to the customer? If there’s not, your model will almost certainly be wrong.