Skip to navigation

RL Framework Compatibility

View as MarkdownOpen in Claude

Reference for NeMo Gym version compatibility with supported training frameworks.

The standalone NeMo Gym image and NeMo RL training images are separate artifacts with independent release schedules. Use nvcr.io/nvidia/nemo-gym to run Gym itself. Use a tested nvcr.io/nvidia/nemo-rl pairing when a NeMo RL recipe requires the training framework and Gym in the same image.


Standalone NeMo Gym Container

Published NeMo Gym release images use the same tag as the Gym release. For example:

docker pull nvcr.io/nvidia/nemo-gym:v0.7.0

Replace v0.7.0 with the release you need after its tag appears in the NeMo Gym NGC catalog. This rule applies to future releases, so this page does not maintain a row for every standalone Gym image.

The existence of a standalone Gym image does not mean that a NeMo RL image bundles the same Gym version.


NeMo RL Training Containers

The following table lists tested NeMo RL pairings. It is not a list of all NeMo Gym releases. NeMo Gym and NeMo RL release independently, so a Gym version can have a standalone image without appearing here.

Match the container to your model recipe, not only your NeMo Gym version. For example, the NeMo RL GRPO tutorial trains Nemotron Nano 9B v2 and uses the Nano NGC container from the v0.1.1 row below—not the v0.3.0 Ultra Dockerfile.

The v0.3.0 row is a historical exception: the Nemotron 3 Ultra pairing has no pre-built NeMo RL image that packages Gym, so it references the ultra-v3 Dockerfile.

For end-to-end setup and training instructions, see the NeMo RL GRPO tutorial.


Unsloth

The NeMo Gym integration with Unsloth is tested on unsloth==2026.1.4 and unsloth_zoo==2026.1.4. Other versions are not guaranteed to work.

For setup and training instructions, see the Unsloth tutorial.


VeRL

NeMo Gym 0.2.1+ is compatible with verl pinned to the commit in REQUIRED_VERL.txt, tested on the verlai/verl:vllm017.latest container (vLLM 0.17.0). Other versions are not guaranteed to work.

For setup and training instructions, see the Training with VeRL tutorial.