extra_loggers is not used to log metrics or hyperparameters

[chajath]

Describe the bug

When I declare a dummy json stdout logger:

from lightning.fabric.loggers import logger
import lightning.pytorch.loggers.logger
from lightning.pytorch.utilities import rank_zero

import simplejson as json


rank_zero_only = rank_zero.rank_zero_only
Logger = lightning.pytorch.loggers.logger.Logger
rank_zero_experiment = logger.rank_zero_experiment


class MyLogger(Logger):
  """Logger override example."""

  @property
  def name(self):
    return "MyLogger"

  @property
  @rank_zero_experiment
  def experiment(self):
    # Return the experiment object associated with this logger.
    pass

  @property
  def version(self):
    # Return the experiment version, int or str.
    return "0.1"

  @rank_zero_only
  def log_hyperparams(self, params):
    print()
    print(json.dumps(params, ignore_nan=True))

  @rank_zero_only
  def log_metrics(self, metrics, step):
    print()
    print(json.dumps({metrics: metrics, step: step}, ignore_nan=True))

And wire them up to the NeMoLogger:

logger = nl.NeMoLogger(ckpt=ckpt, extra_loggers=[MyLogger()])

...
  llm.pretrain(
...
      log=logger,
  )

I don't find json dumps anywhere in the stdout log, suggesting these log methods are not called.

Steps/Code to reproduce bug
Follow the Logger example in the above section and see if the log outputs are printed.

Expected behavior
Called are dumped as jsonl in stdout.

Environment overview (please complete the following information)

• Environment location: GKE
• Method of NeMo install: docker container (tag 24.12)

Environment details
NeMo container 24.12