How to load a pre-trained model

[ccccyf]

I made some changes to “training_module” part after I read the code, but it still didn't seem to work, is there any part i need to write or perhaps I misunderstood something somewhere.? thank u.

training_module:
  _target_: ModelFromCheckpoint
  checkpoint_path: ./transfer_model/best.ckpt

File "/opt/tefscloud/miniconda3/lib/python3.12/site-packages/hydra/_internal/utils.py", line 637, in _locate
    raise ImportError(
ImportError: Error loading 'ModelFromCheckpoint':
ModuleNotFoundError("No module named 'ModelFromCheckpoint'")
Are you sure that module 'ModelFromCheckpoint' is installed?

[MarcD3]

Hello,

There are two things to note:

First, when using a model checkpoint to train from, you still need the training module to be instantiated similarly as to a fresh training run, using something of the form:

training_module:
  _target_: nequip.train.EMALightningModule
  And specify the rest of the training settings how you wish to for this run

Second, to use a model checkpoint for the model instantiation, instead of the model section of the training module being the setup of a nequip or allegro model, it is structured like your code snippet, just change the syntax of ModelFromCheckpoint to nequip.model.ModelFromCheckpoint

The model section within the training module should look like this:

  model:
    _target_: nequip.model.ModelFromCheckpoint
    checkpoint_path: ./transfer_model/best.ckpt

let me know if this helps!

[ccccyf]

Thank you！This way helped me.