Loading custom models

[saninta0212]

Hi all, is there a way for me to load custom models given a path if the model is not present in the model zoo?

[avantikalal]

Yes you can use

from grelu.lightning import LightningModel
model = LightningModel.load_from_checkpoint(file)

However, the model checkpoint has to be saved using gReLU. We currently don't have the ability to load arbitrary pytorch models, though we will add more flexibility in this direction soon.

[saninta0212]

After I load a model for eg using model = grelu.resources.load_model(
project=project_name,
model_name='model'
)

How can I save this model into a checkpoint that can be loaded later?

[avantikalal]

Here is a hacky snippet that works for me, though it isn't ideal.

import pytorch_lightning as pl
trainer = pl.Trainer()
try:
    trainer.predict(model)
except:
    pass
trainer.save_checkpoint(file_path)