Invalid hash code when using EfficientNet

[jdrp]

I am currently on step 3.2 of 06. PyTorch Transfer Learning. When I try to run this code on a Colab notebook:

weights = torchvision.models.EfficientNet_B0_Weights.DEFAULT # .DEFAULT = best available weights 
model = torchvision.models.efficientnet_b0(weights=weights).to(device)

I keep getting the following error:

RuntimeError: invalid hash value (expected "3dd342df", got "7f5810bc96def8f7552d5b7e68d53c4786f81167d28291b21c0d90e1fca14934")

Both the expected and got hash values stay the same if I run the code again.
My torch version is 2.1.0+cu118 and torchvision is 0.16.0+cu118

[mrdbourke]

Hi @jdrp ,

Looks like a few people are getting this issue with the latest version of torchvision.

You can find a solution here: #693 (reply in thread)

In short,

This should help fix it:

Overriding torchvision.models._api.WeightsEnum.get_state_dict() should do the trick, LMK if it doesn't work

from torchvision.models import efficientnet_b0, EfficientNet_B0_Weights
from torchvision.models._api import WeightsEnum
from torch.hub import load_state_dict_from_url

def get_state_dict(self, *args, **kwargs):
    kwargs.pop("check_hash")
    return load_state_dict_from_url(self.url, *args, **kwargs)
WeightsEnum.get_state_dict = get_state_dict

efficientnet_b0(weights=EfficientNet_B0_Weights.IMAGENET1K_V1)
efficientnet_b0(weights="DEFAULT")

Source: pytorch/vision#7744 (comment)

[jdrp]

That worked, thanks for the quick help!

[Aconsumeeverything]

that worked thank you !!