feat: Add recipes

[yfw]

What does this PR do ?

Add nemotron recipes and scripts

Issues

List issues that this PR closes (syntax):

Usage

• You can potentially add a usage example below

# Add a code snippet demonstrating how to use this

Before your PR is "Ready for review"

Pre checks:

• Make sure you read and followed Contributor guidelines
• Did you write any new necessary tests?
• Did you run the unit tests and functional tests locally? Visit our Testing Guide for how to run tests
• Did you add or update any necessary documentation? Visit our Document Development Guide for how to write, build and test the docs.

Additional Information

• ...

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[yfw]

/ok to test 2ee35b4

[macandro96]

Reviewed with a team of six agents (four independent finders across the guide, config, launcher, and tests; two adversarial verifiers that re-derived every claim from source).

Nice work on this — the recipe is complete and self-consistent where it counts. Things that came back clean: all 33 env.nemo_gym.config_paths resolve at the pinned Gym SHA; the 86-node / 344-GPU topology sums correctly and matches the launcher defaults, the YAML header, and both sbatch het components; batch-size and parallelism divisibility all hold (8192 = 512x16, DP=8, EP=16 valid); max_total_sequence_length is consistent across training, generation, and max_model_len; the sbatch argv is well-formed under a stubbed submission; and there are no secrets or internal hostnames. Adding launcher dry-run tests is a genuine improvement over the ultra_launch.sh precedent, which has none.

Nothing below is blocking — some suggestions / nits.

Generated by Claude Code

[yfw]

/ok to test cf68648