When you're evaluating a model on HuggingFace, you usually have to dig into the model card, check the repo files, or run the model locally just to find out basic architecture details how many layers, what's the hidden size, what precision it runs at.
