> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.nvidia.com/nemo/gym/main/orchestration/how-to/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.nvidia.com/nemo/gym/_mcp/server.

# How-To Guides

> Task-oriented guides for gym eval submit: multi-node vLLM, prefill/decode, node pools, custom harnesses, and resumable runs.

Each guide builds on the [Quickstart](/orchestration/quickstart) and shows only the parts of the config that change. Field details are in the [Configuration Reference](/orchestration/configuration).

#### [Multi-Node vLLM](/orchestration/how-to/multi-node-vllm)

Run many replicas, or one very large instance, across nodes, and learn when the Ray Serve gateway takes over.

#### [Prefill/Decode Disaggregation](/orchestration/how-to/prefill-decode-disaggregation)

Serve a model as separate prefill and decode tiers behind a router.

#### [Node Pools and Ray](/orchestration/how-to/node-pools-and-ray)

Give a judge or scorer nodes of its own, and span a Ray cluster across pools.

#### [Custom Harness](/orchestration/how-to/custom-harness)

Run your own script instead of `gym eval run`.

#### [Resumable Runs](/orchestration/how-to/resumable-runs)

Continue automatically after a time limit, preemption, or node failure.

> Task-oriented guides for gym eval submit: multi-node vLLM, prefill/decode, node pools, custom harnesses, and resumable runs.

## Docs

- [Multi-Node vLLM](https://docs.nvidia.com/nemo/gym/main/orchestration/how-to/multi-node-vllm.md): Scale a vLLM service across GPUs and nodes with gym eval submit.
- [Prefill/Decode Disaggregation](https://docs.nvidia.com/nemo/gym/main/orchestration/how-to/prefill-decode-disaggregation.md): Serve a model as separate prefill and decode vLLM tiers behind a vllm-router.
- [Node Pools and Ray](https://docs.nvidia.com/nemo/gym/main/orchestration/how-to/node-pools-and-ray.md): Give services nodes of their own and span a Ray cluster across node pools.
- [Custom Harness](https://docs.nvidia.com/nemo/gym/main/orchestration/how-to/custom-harness.md): Run your own script instead of gym eval run for a benchmark with its own harness.
- [Resumable Runs](https://docs.nvidia.com/nemo/gym/main/orchestration/how-to/resumable-runs.md): Continue a benchmark automatically after a time limit, preemption, or node failure.