Error when instantiating Logger with **kwargs only arguments via CLI (NeptuneLogger)

[verginer]

I am running a model via the CLI interface and instantiating the Logger class using the class_path and init_args format. It works well when the parameters are explicitly defined in the __init__ method, but it does not work for parameters that are only provided as keyword arguments (kwargs). In the case of the NeptuneLogger, the neptune_run_kwargs parameters such as tags raise an error during the instantiation of the logger.

What version are you seeing the problem on?

v2.0

How to reproduce the bug

python train.py fit -c config/model.yaml

The model.yaml file looks like this:

trainer:
    logger:
        class_path: lightning.pytorch.loggers.NeptuneLogger
        init_args:
            project: "<project/name>"
            tags:
                - "mlp"
                - "fast-dev"

Error messages and logs

train_embedseq.py: error: Parser key "trainer.logger":
  Does not validate against any of the Union subtypes
  Subtypes: (<class 'lightning.pytorch.loggers.logger.Logger'>, typing.Iterable[lightning.pytorch.loggers.logger.Logger], <class 'bool'>, <class 'NoneType'>)
  Errors:
    - Problem with the given class_path 'lightning.pytorch.loggers.NeptuneLogger':
        'Configuration check failed :: No action for destination key "tags" to check its value.'
    - Expected a <class 'collections.abc.Iterable'>
    - Expected a <class 'bool'>
    - Expected a <class 'NoneType'>
  Given value type: <class 'jsonargparse.namespace.Namespace'>
  Given value: Namespace(class_path='lightning.pytorch.loggers.NeptuneLogger', init_args=Namespace(api_key=None, project='verginer/nlpatent', name=None, log_model_checkpoints=True, prefix='mlp', tags=['embed-seq']))

Environment

No response

More info

No response

[diegoluna3]

Also getting a similar issue. I tried specifying dict_kwargs like the suggestion from this WandbLogger issue, but that also did not work:

OS

Ubuntu 20.04.6

Versions

neptune 1.3.1
lightning 2.0.6
jsonargparse 4.23.0
python 3.11.4

YAML

trainer:
  accelerator: gpu
  devices: 1
  logger:
    class_path: lightning.pytorch.loggers.NeptuneLogger
    init_args:
      api_key: ${NEPTUNE_API_TOKEN}
      project: ${NEPTUNE_PROJECT}
      log_model_checkpoints: true
      prefix: training
      dict_kwargs:
        tags:
          - 'model2'
          - 'binary'

Traceback

main.py: error: Parser key "trainer.logger":
  Does not validate against any of the Union subtypes
  Subtypes: (<class 'lightning.pytorch.loggers.logger.Logger'>, typing.Iterable[lightning.pytorch.loggers.logger.Logger], <class 'bool'>, <class 'NoneType'>)
  Errors:
    - Problem with given class_path 'lightning.pytorch.loggers.NeptuneLogger':
        'Configuration check failed :: No action for destination key "dict_kwargs.tags" to check its value.'
    - Expected a <class 'collections.abc.Iterable'>
    - Expected a <class 'bool'>
    - Expected a <class 'NoneType'>
  Given value type: <class 'jsonargparse._namespace.Namespace'>
  Given value: Namespace(class_path='lightning.pytorch.loggers.NeptuneLogger', init_args=Namespace(api_key='${NEPTUNE_API_TOKEN}', project='${NEPTUNE_PROJECT}', name=None, log_model_checkpoints=True, prefix='training', dict_kwargs=Namespace(tags=['model2', 'binary'])))

[diegoluna3]

@verginer I realized there was an indentation mistake in my yaml file. If you try this it should work:

trainer:
    logger:
        class_path: lightning.pytorch.loggers.NeptuneLogger
        init_args:
            project: "<project/name>"
        dict_kwargs:
            tags:
                - "mlp"
                - "fast-dev"