huggingface/pytorch-image-models

No pretrained weights exist

FeU-aKlos opened this issue · 1 comments

FeU-aKlos commented

Describe the bug
RuntimeError: No pretrained weights exist for efficientnetv2_m. Use pretrained=False for random init.

To Reproduce
Steps to reproduce the behavior:
import timm
model = timm.create_model("efficientnetv2_m",pretrained=True)

Desktop (please complete the following information):

  • OS: Ubuntu 22.04
  • timm==0.9.2
rwightman commented

No weights exist for that model as per the message. It's a pytorch native variant of the model with normal padding. The tensorflow ported weights w/ tf 'SAME' padding are tf_efficientnetv2_m.in21k_ft_in1k and tf_efficientnetv2_m.in1k

There is also my 'm' config with more differences than just padding efficientnetv2_rw_m.agc_in1k ..

Please look at the huggingface hub or pretrained_cfg in model code for other variants.

  • 👍1