How to solve the bug"OSError: [Errno 38] Function not implemented" when I use the train.py?

[Calvin-Pang]

Hello,I am trying to use the train.py in Colab. And I use the command !python train.py /content/drive/My\ Drive/pytorch-image-models/imagenette2-320 --model resnet34
But when it seems to complete the training process, the bug exists:
### Traceback (most recent call last):
File "train.py", line 799, in
main()
File "train.py", line 618, in main
best_metric, best_epoch = saver.save_checkpoint(epoch, metric=save_metric)
File "/content/drive/My Drive/pytorch-image-models/timm/utils/checkpoint_saver.py", line 78, in save_checkpoint
os.link(last_save_path, save_path)
OSError: [Errno 38] Function not implemented: '/content/drive/My Drive/pytorch-image-models/output/20210705-131030-resnet34-224/last.pth.tar' -> '/content/drive/My Drive/pytorch-image-models/output/20210705-131030-resnet34-224/checkpoint-0.pth.tar'
Is anyone can help me?

[amaarora]

Have you tried to update the package? !pip install timm --upgrade

[Calvin-Pang]

Thank you for the advice. However, I tried this command in Colab and it told me that the timm-0.4.12 is already the latest version. And I tried to run train.py again, the bug still exists.

[rwightman]

@Calvin-Pang current checkpoint saver only works on filesystems that support hard links, which is most filesystems except for some like those mounted via fuse, some network fs (NFS, SAMBA w/ Unix ext should work), and old windows filesystems.

I'm taking that into consideration for the next iteration of training scripts (in bits branch), see https://github.com/rwightman/pytorch-image-models/blob/bits_and_tpu/timm/bits/checkpoint_manager.py#L84-L104 ... but will not be adding support to the existing one.

[Calvin-Pang]

Thank you very much for your reply. I will try it in other filesystems.