resume training

[ireneb612]

Hi When I restart the training from a previous checkpoint, as soon as the scripts tries to save the new checkpoint it throws an error!

Traceback (most recent call last):
File "train.py", line 802, in
main()
File "train.py", line 622, in main
best_metric, best_epoch = saver.save_checkpoint(epoch, metric=save_metric)
File "/content/drive/MyDrive/Compact-Transformers/timm/utils/checkpoint_saver.py", line 78, in save_checkpoint
os.link(last_save_path, save_path)
OSError: [Errno 38] Function not implemented: './output/train/20220210-115030-cct_7_3x1_32_c100-32/last.pth.tar' -> './output/train/20220210-115030-cct_7_3x1_32_c100-32/checkpoint-0.pth.tar'

It seam like there is a problem with the paths!!

[rwightman]

@ireneb612 the checkpoint saver doesn't work on filesystems that don't support hard linking ... colab would be one of those situations. bits_and_tpu branch (a collection of ideas for next iteration of timm training scripts, etc, and TPU support) includes a workaround but I don't plan on adding it to the current master

[ireneb612]

Thank you for your answer I appreciate your work!