NVIDIA-NeMo / NVIDIA-NeMo/RL

Flag name should_log_nemo_gym_responses is confusing

Open
#2,472 0 comments 1 reaction 1 assignee Claimed by @terrykong View on GitHub
bug
Dominant language
Python
Stars
2k
Forks
561
Avg merge
4d 5h
Merged PRs (30d)
145

Description

When running GRPO with NeMo-Gym, I set:

```yaml
env:
should_use_nemo_gym: true
should_log_nemo_gym_responses: false
```

I expected this to reduce logging/storage usage. Instead, NeMo-RL writes large files like:

```text
/exp_002/train_data_step1.jsonl
```

This seems to happen because `grpo.py` writes train data when the flag is false:

```python
if not _should_log_nemo_gym_responses(master_config):
logger.log_batched_dict_as_jsonl(
log_data, f"train_data_step{total_steps + 1}.jsonl"
)
```

This is confusing because the config name sounds like it only controls NeMo-Gym response logging.

Could we add a separate config knob to disable these files, e.g.:

```yaml
logger:
log_train_data_jsonl: false
```

or rename/split the existing flag so low-storage logging is clearer?

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.