cannot pickle ‘nvidia.dali.backend_impl.tfrecord.Feature’ object

[PYNing]

I'd like to integrate redoxify-dataloader to my config by adding code

import sys
sys.path.append('/workspace/github/redoxify-dataloader/src')
sys.path.append('/workspace/github/redoxify-dataloader/')
custom_imports = dict(
    imports=['redoxify.plugin.mmdetection.runner.RedoxMMYoloRunner'],
    allow_failed_imports=False)
from examples.mmtrain.redox_config import redox_dataset_config as redox_dataset_config
runner_type='redoxify.plugin.mmdetection.runner.RedoxMMYoloRunner'

but the exception above was raised when running

# train.py
cfg = Config.fromfile(args.config) 

Running example is OK in the same version of MMEngine. Is something missing?

[PYNing]

Supplement: The root cause of the problem is that when the config contains "base", the following deepcopy() will be triggered.

https://github.com/open-mmlab/mmengine/blob/main/mmengine/config/config.py#L795