How to predict a system's energies through a pt file? 🙏

[The-one-on-the-moon]

I was trying to verify a pt file's accuracy which is finetuned from dpa-2.
Then I met an issue about predicting energies through a pt file.

training_systems = dpdata.LabeledSystem(path, fmt='deepmd/npy')

# predict = training_systems.predict('./frozen_model.pth')
predict = training_systems.predict('./model.ckpt.pt')

The error was below

 File "~/deepmd-kit/lib/python3.11/site-packages/deepmd/infer/deep_eval.py", line 189, in _load_graph
    graph_def.ParseFromString(f.read())
google.protobuf.message.DecodeError: Error parsing message with type 'tensorflow.GraphDef'

I've done this prediction by a pb file, so I wonder if it works with pt file.

Thanks for your support, looking forward to your replies !

[njzjz]

Do you use DeePMD-kit v3?

[The-one-on-the-moon]

Yes DeePMD-kit v3.0.0b2

[njzjz]

Actually, the code you posted is not v3.0.0b2

File "~/deepmd-kit/lib/python3.11/site-packages/deepmd/infer/deep_eval.py", line 189, in _load_graph
graph_def.ParseFromString(f.read())

[The-one-on-the-moon]

I apologize for my mistake. I checked my environment and realized I was not using v3 as you suggested. After switching to v3, everything worked perfectly. Thanks a lot for your help!