> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.nvidia.com/nemo/gym/llms.txt.
> For full documentation content, see https://docs.nvidia.com/nemo/gym/llms-full.txt.

# Success Criteria

Use these criteria to validate that your Gym integration is working correctly. A successful integration must pass all validation benchmarks.

<Tip>
  These success criteria may evolve as new integration challenges are discovered. Check this page for updates when troubleshooting integration issues.
</Tip>

## Validation Checklist

### 1. Component Form Factor

Verify that your integration implements all required components as specified in [Gym Integration Footprint And Form Factor](/latest/contribute/rl-framework-integration/gym-integration-footprint-and-form-factor):

* [ ] OpenAI-compatible HTTP server
* [ ] On-policy token ID fixes
* [ ] Gym spinup and integration
* [ ] Rollout orchestration
* [ ] Training loop integration

### 2. Environment Configuration

Verify that your integration can load and run arbitrary Gym training environments through configuration:

* [ ] Environment configuration loads from YAML
* [ ] Multiple environments can be selected at runtime
* [ ] Environment parameters are configurable without code changes

### 3. Math Reasoning Benchmark

Train on the DAPO17k math training environment and verify model improvement on AIME24.

| Parameter              | Value                                                                                                                                                                             |
| ---------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Training environment   | [DAPO17k math environment](https://github.com/NVIDIA-NeMo/Gym/blob/299e8c04f4a3bbf0f6069139092225f2fe3aa70f/resources_servers/math_with_judge/configs/bytedtsinghua_dapo17k.yaml) |
| Base model             | [Qwen3-4B-Instruct-2507](https://huggingface.co/Qwen/Qwen3-4B-Instruct-2507)                                                                                                      |
| Minimum training steps | 1,000                                                                                                                                                                             |
| Validation set         | AIME24 (included with training environment)                                                                                                                                       |
| Target accuracy        | ≥85%                                                                                                                                                                              |

### 4. Workplace Assistant Benchmark

Train on the workplace assistant environment and verify validation set improvements.

| Parameter              | Value                                                                                                                                                     |
| ---------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Training environment   | [Workplace assistant environment](https://github.com/NVIDIA-NeMo/Gym/tree/299e8c04f4a3bbf0f6069139092225f2fe3aa70f/resources_servers/workplace_assistant) |
| Base model             | [Qwen3-4B-Instruct-2507](https://huggingface.co/Qwen/Qwen3-4B-Instruct-2507)                                                                              |
| Minimum training steps | 100                                                                                                                                                       |
| Success criterion      | Observable validation set improvement                                                                                                                     |

## Troubleshooting

If your integration fails to meet the success criteria:

1. **Training crashes**: Check for off-policy issues. Refer to [On-Policy Corrections](/latest/contribute/rl-framework-integration/openai-compatible-http-server-on-policy-correction)
2. **No improvement**: Verify rollout orchestration is correctly tracking token IDs
3. **Environment errors**: Verify OpenAI-compatible HTTP server endpoints match the specification

## Related Topics

* [Gym Integration Footprint And Form Factor](/latest/contribute/rl-framework-integration/gym-integration-footprint-and-form-factor) - Required integration components
* [On-Policy Corrections](/latest/contribute/rl-framework-integration/openai-compatible-http-server-on-policy-correction) - On-policy training fixes