Confused with the idea that make downloads as an indicator to reflect how the candidate models suit the issue.

[Pied-Piper1]

"To some extent, we think the downloads can reflect the popularity and quality of the model.". This is what I extract from your published paper. I wonder if it's precise enough to say that the downloads can reflect the popularity and quality of the model? Or you're unable to find a more convincing and precise way to filter out models?