Hello @mike-dd,
Please, can you help upgrade Timm package on your end to the latest version.
My trained model is unable to initialise since it wasn’t captured in the version of Timm currently on the submission backend.
The latest version is 0.6. It can be installed directly from official repo: !pip install git+https://github.com/rwightman/pytorch-image-models
Thank you.
@flamethrower I think you can clone it to your submission folder and append it to the sys path instead of waiting to update the version. i didn’t try it but it should work
1 Like
Yes thank you, I can easily just include the repo in my submission.
However, @mike-dd, please can you confirm it is alright to include pre-trained models available on the latest version of timm: !pip install git+https://github.com/rwightman/pytorch-image-models.
It is publicly available to everyone and timm package is originally included in the installed modules on backend. I just want to use a pre-trained regnet model.
Thank you for your confirmation.
Hi @flamethrower,
can you confirm it is alright to include pre-trained models available on the latest version of timm
Yes, that is fine.
I’m glad to hear that Ammarali32’s suggested workaround is doable for you. We would rather not update timm
to the pre-release 0.6.0
version, particularly as the 0.5.4
release notes indicates there are some riskier changes coming up.
I hope this helps. Feel free to keep us posted.