Find better training recipes through reproducible experiments.
What it does
Runs iterative reinforcement-learning experiments and records their results to improve the user's chosen training metric.
When to use it
Use for research campaigns that test hypotheses, improve recipes, or pursue open-ended discovery with NeMo-RL or NeMo-gym.
How to use it
Give it a research objective in the repository; it runs experiments and records hypotheses and results in branches and logs.
What you provide
- An existing project
What you need · 5
The prescribed experiment command uses uv to run the selected entrypoint.
GPU experiments need suitable GPU capacity in the chosen local or remote execution environment; CPU-only local runs are restricted to light checks.
Required for every campaign to start or resume a session record and checkpoint plans, edits, launches, steering, and handoffs.
Access · 1
This skill
NeMo-RL
Launches long-running training jobs