Open BryanWBear opened 3 weeks ago
@BryanWBear Yes! I am turning this ticket into a feature request, which i can work on it later. Thanks for bringing this up.
For now, to save your reft model, you can also try reft_model .save(<your_dir>)
to save by using our own API instead of the trainer's API. I think this API will save the config as well as other artifacts.
The issue is that
ReftTrainer.save_model
does not save theReftConfig
, only the intervention.As a workaround, we can load the model from the checkpoint using the following code (by reinstantiating the config manually):
Please let me know if I am missing something!
Thanks, Bryan