What's the difference of the 4 models?

[deepfree2023]

Hi, is there any guidance on how to choose among the 4 models?

[WeirdScienceX]

https://huggingface.co/microsoft/Florence-2-base

[deepfree2023]

Thanks, I've read this before use.

But the actual performance of the models seems quite different from the description.

In my tests, the -ft ones seems always produce worse results on every task compared to non-ft ones.

[kijai]

I have heard that one possible reason for the -ft variants being worse, is that they are finetuned to do well in benchmarks. I really don't know how true that is though.