timm
/

Image Classification
timm
PyTorch
Safetensors

Showing width vs depth of each model

#1
by liuyao - opened

It may be a feature request, not just for timm models but huggingface in general. I wanted to see how a model processes the input as you progress through the layers, and here's what I came up with for resnet50 (any error is on me). I'm most curious to compare different models, and I think it would give a glimpse of the model on the model card, without understanding all the details. (I would think someone must have done this before.)

width_resnet50.jpg

I made it so that the area under the graph should tell you the number of parameters in use, but this is in log scale. What do you think?

Sign up or log in to comment