Skip to content

Fix super3 data prep rl subgroup bootstrap#269

Open
fallintoplace wants to merge 1 commit into
NVIDIA-NeMo:mainfrom
fallintoplace:fix/super3-data-prep-rl-subgroup
Open

Fix super3 data prep rl subgroup bootstrap#269
fallintoplace wants to merge 1 commit into
NVIDIA-NeMo:mainfrom
fallintoplace:fix/super3-data-prep-rl-subgroup

Conversation

@fallintoplace

Copy link
Copy Markdown

Summary

  • register super3 data prep rl as a Typer subgroup instead of importing the shadowed prep.rl module surface
  • remove the obsolete src/nemotron/cli/commands/super3/data/prep/rl.py dispatcher and add explicit rlvr, swe1, swe2, and rlhf data-prep command modules under the rl/ package
  • add a CLI regression test and update the Super3 docs/comments that still advertised the flat super3 data prep rl entrypoint

Root cause

src/nemotron/cli/commands/super3/data/prep/_typer_group.py imported META and rl from nemotron.cli.commands.super3.data.prep.rl, but that path now has both a sibling rl.py file and an rl/ package. Python resolves the package first, and the package only exported rl_app, so the Super3 data-prep CLI broke during import/bootstrap.

Impact

After this change, super3 data prep rl loads as the intended subgroup and exposes all documented stage-specific subcommands:

  • rlvr
  • swe1
  • swe2
  • rlhf

Validation

  • python3 -m compileall src/nemotron/cli/commands/super3/data/prep tests/recipes/super3/test_cli.py
  • uv run --no-project --with pytest --with "nemo-run @ git+https://github.com/NVIDIA-NeMo/Run.git@main" --with typer --with rich --with textual --with fsspec --with numpy --with pyarrow --with xxhash --with "transformers>=4.57.6,<5.0" --with "huggingface_hub>=0.20.0" --with "datasets>=2.14.0" --with pyyaml --with tomlkit --with colorama --with "wandb>=0.23.1" --with jinja2 --with pydantic --with "pydantic-settings>=2.12.0" --with omegaconf env PYTHONPATH=src pytest tests/recipes/super3/test_cli.py

@chrisalexiuk-nvidia

Copy link
Copy Markdown
Contributor

Need signoff.

(cherry picked from commit 7e57c65)
Signed-off-by: Minh Vu <vuhoangminh97@gmail.com>
@fallintoplace
fallintoplace force-pushed the fix/super3-data-prep-rl-subgroup branch from b877002 to 4e12cec Compare July 8, 2026 18:46
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants