How do I save the model locally

[rishiwobb]

Describe the bug
I wish to save a pre-trained model after calling the create_model method and then later on load the model from the file locally. How to do it?

[hankyul2]

hi @rishiwobb

When you download the pre-trained weight, you can find the download path which is usually the ~/.cache/torch/hub/checkpoints folder. Next, you can manually load the already downloaded pre-trained model like below code snippet.

import timm
model = timm.create_model('resnet50', checkpoint_path='/home/hankyul/.cache/torch/hub/checkpoints/resnet50_a1_0-14fe96d1.pth')

Thank you.

hankyul

[hamid-mp]

Hi,@hankyul2
as you mentioned, is this the best way to save and load our model using timm?

1- Build Model:

timm.create_model(model_name='resnet50d',
                          pretrained=True,
                          num_classes=84,
                          in_chans=3) 

2- Save fine-tuned Model:

torch.save(model.state_dict(), f='custom_weights.pt' 

3- Load Model:

        timm.create_model(model_name='resnet50d',
                          checkpoint_path=custom_weights.pt,
                          num_classes=84,
                          in_chans=3) 

[hankyul2]

Hi @hamid-mp

I'm sorry for not getting back to you sooner. Yes, you are right. I suggest just loading the downloaded weight, not saving and loading the finetuned weight.

Thank you for your correction.

Hankyul.