Skip to content

fix issue #8 add code for continue training option

Add some minor code refactoring to implement the continue training option.

For the moment the continue training option is only implemented to resume an interrupted training session and not to be used for fine-tuning and using pre-trained model.

Some minor refactoring has been made in Odeon to implement this functionality :

  • move/centralize save_model and load_model functions to the module models.py
  • made a new function to define all filenames/pth associated to training a model (model.pth, optimizer_model.pth model_history.json ...)
  • add a check in Train tools to test if models files exists when continue training is activated.

TODO :

  • update doc
Edited by nicolas david

Merge request reports