Saving the trained model and used it in other code

[karfungc]

Is there a method to saving the trained PINN model and call it in another code?

[dario-coscia]

Hello👋🏼 There are methods to save a trained model, solver and trainer. By default Pytorch Lightining creates loggers and checkpoints during training saving:

• 16-bit scaling factor (if using 16-bit precision training)
• Current epoch
• Global step
• LightningModule’s state_dict (PINA Solver)
• State of all optimizers
• State of all learning rate schedulers
• State of all callbacks (for stateful callbacks)
• State of datamodule (for stateful datamodules)
• The hyperparameters (init arguments) with which the model was created
• The hyperparameters (init arguments) with which the datamodule was created
• State of Loops

more information here

By default all checkpoints for each run are saved in the current directory under a directory called lightning_logs. If you want to specify where to save the checkpoints just passed the argument default_root_dir to the Trainer before training. Then for restoring you can either restore training: trainer.train(ckpt_path=[YOUR PATH]); or you can restore a trained solver: trained_solver = SOLVER.load_from_checkpoint( [YOUR PATH], problem, model) where SOLVER is the class object representing any pina solver (e.g. PINN, SupervisedSolver, …)

For a working example look at the test_train_restore and test_train_load method in the tests folder .

Let me know how it goes and if it was helpful for you😃. If you like the package leave us a star ⭐️ which helps us grow the community!

[karfungc]

Thank you! Will check it