Trained checkpoint from Pointcept missing 'config' key when loading

Hi, I trained a model using Pointcept with the following command:
sh scripts/train.sh -m 1 -g 8 -d sonata -c semseg-sonata-v1m1-3a-s3dis-lin -n semseg-sonata-v1m1-0-base-3a-s3dis-lin -w exp/sonata/pretrain-sonata-v1m1-0-base/model/model_last.pth
After training, I tried to load the checkpoint using:
model = sonata.load("/root/pythoncode/Pointcept-main/exp/sonata/semseg-sonata-v1m1-0-base-3a-s3dis-lin/model/model_best.pth").cuda()
But I encountered this error:
KeyError: 'config'
It seems that the 'config' key is missing from the checkpoint file.

Could you please help me with the following:

How can I include the config when saving the model during training?

Alternatively, how can I manually reconstruct the config used for training this model?

Thanks in advance!

Start by tracing the checkpoint produced by scripts/train.sh and the loading path used by sonata.load(.../model_best.pth), then compare the saved keys with the training configuration supplied in the command. Done means establishing whether checkpoints should contain config and documenting or testing a reliable way to load this trained checkpoint without the KeyError.

Written by the indexing model from the issue text.