PicklingError

[kghezelbash]

Hi again =)

I was fine-tuning the 'crnn-fa-custom' yesterday with no problem, but when I ran the same code today it raised this error at the end of the first epoch:

[/usr/local/lib/python3.10/dist-packages/hezar/trainer/trainer.py](https://localhost:8080/#) in save(self, path, config_filename, model_filename, model_config_filename, subfolder, dataset_config_file, optimizer_file)
    722         self.config.save(path, filename=config_filename, subfolder=subfolder)
    723         self.model.save(path, filename=model_filename, config_filename=model_config_filename)
--> 724         torch.save(self.optimizer, os.path.join(path, subfolder, optimizer_file))
    725         if isinstance(self.train_dataset, Dataset):
    726             self.train_dataset.config.save(path, filename=dataset_config_file, subfolder=subfolder)

[/usr/local/lib/python3.10/dist-packages/torch/serialization.py](https://localhost:8080/#) in save(obj, f, pickle_module, pickle_protocol, _use_new_zipfile_serialization, _disable_byteorder_record)
    626     if _use_new_zipfile_serialization:
    627         with _open_zipfile_writer(f) as opened_zipfile:
--> 628             _save(obj, opened_zipfile, pickle_module, pickle_protocol, _disable_byteorder_record)
    629             return
    630     else:

[/usr/local/lib/python3.10/dist-packages/torch/serialization.py](https://localhost:8080/#) in _save(obj, zip_file, pickle_module, pickle_protocol, _disable_byteorder_record)
    838     pickler = pickle_module.Pickler(data_buf, protocol=pickle_protocol)
    839     pickler.persistent_id = persistent_id
--> 840     pickler.dump(obj)
    841     data_value = data_buf.getvalue()
    842     zip_file.write_record('data.pkl', data_value, len(data_value))

PicklingError: Can't pickle <class '__main__.PersianOCRDataset'>: it's not the same object as __main__.PersianOCRDataset

Has anything changed that is related to fine-tuning?

[arxyzan]

Hi @kghezelbash . The latest release had some big changes in the trainer regarding checkpointing and all tests were passed.
Can you please share a reproducible code that causes this error?
EDIT: Use the source package instead of the latest release please (pip install git+https://github.com/hezarai/hezar)

[kghezelbash]

Thank you! It is working with pip install git+https://github.com/hezarai/hezar now =))))

[arxyzan]

@kghezelbash Thanks to you since there was actually a bug in the Trainer in v0.38.0 which is now fixed in the latest version (v0.38.1)
So you can still use the regular pip install hezar now.

[kghezelbash]

Thanks, I just test the new version and it is working =))