Batch Serving
The Batch Serving workload umbrella defines recommended, cohesive deployments for processing large-scale, offline, or latency-insensitive tasks on llm-d infrastructure.
Depending on your integration requirements, scale, and operational environment, llm-d offers two distinct paths for queue-based and batch inference:
- Batch Gateway: An enterprise-grade, fully managed OpenAI-compatible Batch API (
/v1/batches,/v1/files). Best for multi-tenant environments where clients require formal asynchronous job submission, file storage, status tracking, and strict separation of interactive vs. batch compute. - Asynchronous Processing: A lightweight, low-overhead queue dispatch mechanism (using Redis or GCP Pub/Sub). Best for internal microservice architectures that require low-complexity background task processing or for filling "slack" capacity in your inference pool via dynamic dispatch gating.