hyper parameters not restored while resuming training

[Haojia521]

I call save_hyperparameters() in __init__(), and all hyper parameters sent to PL model are saved to checkpoint file. However, when i resume training from a checkpoint(call trainer.fit(..., ckpt_path=checkpoint_file_path)), the hyper parameters are not restored from checkpoint file and all of them keep initial values.

[rohitgr7]

hyperparameters are not restored by default because it allows users to update them if they want, using the checkpoint, while resuming.

you can do this:

model = LitModel.load_from_checkpoint(checkpoint_file_path)
trainer.fit(model, ..., ckpt_path=checkpoint_file_path)

[Haojia521]

ok, it works fine. Thanks very much for your advice.