Integration with nemo-gym

[eligotts]

Description

Integration with nemo-gym. Look in environments for example integrations. Key decision: spin up one sandbox per environment and run the resource server in the sandbox.

Tests:

Type of Change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Documentation update
• Test improvement

Testing

• All existing tests pass when running uv run pytest locally.
• New tests have been added to cover the changes

Checklist

• My code follows the style guidelines of this project as outlined in AGENTS.md
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• Any dependent changes have been merged and published

Additional Notes

[cdreetz]

can probably create an environments/nemogym/nemo_* to organize them all a bit instead of being flat in environments/

same thing in verifiers/integrations/ can create a verifiers/integrations/nemogym/nemo_gym_env.py and verifiers/integrations/nemogym/utils.py or something of the sorts

[cmunley1]

This is super cool! I think that this definitely works as a first step to unlock many, but not all, environments in NeMo Gym, through verifiers.

Some nemo gym environments control the multi turn loop e.g. through agents, so for a next step we can explore something like this: https://github.com/PrimeIntellect-ai/verifiers/blob/main/verifiers/envs/experimental/rollout_gateway_mixin.py

I will try to prepare a draft to share.