Guides#

Use Guides for operational workflows: configure a run, launch it at scale, manage data and checkpoints, and tune performance.

Choose a Workflow#

Configure

Set up Hydra configs, logging, and workload-specific options.

Launch & Scale

Run across nodes, heterogeneous hardware, cloud-edge setups, or robots.

Data & Checkpoints

Collect data, convert checkpoints, and resume training.

Performance

Tune placement, scheduling, parallelism, profiling, and adapters.

Hardware Backends

Run RLinf on AMD ROCm or Ascend CANN.

Agent Workflows

Operate agentic and reasoning workloads.

Rollout Engines

Launch sglang servers and routers behind a single inference endpoint.