GenRM Service API

The GenRM (Generative Reward Model) service provides LLM-based response evaluation. It is deployed as a Ray Serve deployment with a FastAPI ingress.

Overview

Property	Value
Module	relax.components.genrm
Deployment	@serve.deployment(logging_config=...)
Ingress	FastAPI

Architecture

Unlike Actor and Rollout, GenRM is a passive HTTP service — it does not run a background loop. It only responds to incoming /generate requests.

The service uses SGLang engines to perform preference evaluation:

1. Receives OpenAI-format chat messages via /generate
2. Applies chat template and tokenizes the prompt
3. Sends to SGLang engine with configurable sampling parameters
4. Returns raw model response text

Colocated Mode

When colocated with the Actor (sharing GPU resources), GenRM supports offload/onload operations:

• Offload: Releases GPU memory before Actor training
• Onload: Loads model weights back to GPU before rollout

Two colocate sub-modes are auto-detected from the GPU allocation:

• Split (rollout_num_gpus + genrm_num_gpus == actor_total_gpus): GenRM and Rollout occupy disjoint bundles.
• Shared (rollout_num_gpus == genrm_num_gpus == actor_total_gpus): GenRM and Rollout occupy the same bundles, splitting each GPU's memory via SGLang mem_fraction_static. GenRM reads its mem_fraction_static from --genrm-engine-config. GenRM never sources weights from the Actor; onload only resumes its KV cache and CUDA graphs.

See GenRM example for full configuration.

Multiple Instances (route_key)

A single GenRM deployment can host several independent judge models at once (configured via --genrm-instances), selected per request via the route_key field in the request body. Omitting route_key (or running the legacy single-instance --genrm-model-path config) routes to the sole "__default__" instance. For the config format, per-instance /health / /metrics detail, and an agentic module-routing example, see GenRM Example · Multi-Instance GenRM.

HTTP Endpoints

Source

• Implementation: relax/components/genrm.py
• Base class: relax/components/base.py