PL with Timm gives KeyError: 'model.model.conv1.0.bias when calling trainer.test

[MaxPolak97]

Hi, I'm using Timm for creating Resnet model. For training it works fine but when calling trainer.test(light_model, datamodule, ckpt_path='best') I get the following error KeyError: 'model.model.conv1.0.bias'. I have made three classes TransferLearningModel for loading any pre-trained model, DataModule for loading data and LightningClassifier model to do the actual training. So I checked how the model was defined and see that it goes wrong at .0 (see below). I didn't has this problem when using torchvision.models!

However, when I use [0] instead of .0, it works.

Does anyone know how this problem occurred and can help me with solving this?

[akihironitta]

@MaxPolak97 Would you mind providing a minimal reproducible script and your env details to investigate this?