Rollout Engines

Use these guides to bring up the inference engines RLinf rollouts talk to over HTTP — server-side processes you launch alongside (or independently of) your training run, plus the client used to call them.

Guide	What you get
SGLang Server & Router	Launch an sglang HTTP server group and an sglang router, with a single OpenAI-compatible endpoint for /generate and /v1/chat/completions.
Calling SGLang with InferenceHTTPClient	Send sync and async /generate / /v1/chat/completions requests to a router (or a single server) from your own code.
SGLang Version Switching	Switch between SGLang versions for the rollout engine.