Is there an easy way to save checkpoints for Colab user? about renet

[dahyun-kang]

Hello,

This code does not provide a resume option at this point because training RENet takes a few hours so I used to wait and see 😃.

You may want to use the -save_all option for the first run and load_model function for the subsequent run to load the last checkpoint that you saved in the first run. Also, the optimizer should be properly loaded as well, which the official code does not provide.

Below are the code snippets you might want to refer to:

• 🔽 Use the -save_all argument for the first run.
  

  renet/common/utils.py

  Line 149 in ee315c1

  parser.add_argument('-save_all', action='store_true', help='save models on each epoch')

• 🔽 Call the load_model function right after this line below so that the checkpoint is properly loaded to the model.
  

  renet/train.py

  Lines 92 to 93 in ee315c1

  	model = RENet(args).cuda()
  	model = nn.DataParallel(model, device_ids=args.device_ids)

• 🔽 Here is the load_model function in common/utils.py.
  

  renet/common/utils.py

  Line 76 in ee315c1

  def load_model(model, dir):

Have a great day! 😃