vss-deploy-dense-captioning

To deploy a standalone RT-VLM dense-captioning service, use the provided Docker Compose workflow with the bp_developer_alerts_2d_vlm profile. It handles preflight checks, model loading, and endpoint health verification to ensure a repeatable local validation environment.

Available RT-VLM dense-captioning API endpoints include /v1/files, /v1/generate_captions, /v1/streams/add, and /v1/chat/completions. OpenAPI-driven endpoint discovery validates these routes on port 8018 during the deployment workflow.

Yes, GPU-enabled deployment is required to run RT-VLM dense-captioning locally. You also need valid NGC and Kafka credentials to handle model loading and wire the Kafka topics for caption and incident streaming.

Yes, RT-VLM deployment supports Kafka-based caption and incident streaming. The deployment workflow includes topic wiring and health checks to validate Kafka output alongside the REST API endpoints.

The best way to validate RT-VLM endpoint health and model loading is through the documented deployment workflow. It includes preflight checks, API-centric validation, and error handling to ensure the service is fully operational on port 8018.

RT-VLM dense-captioning deployment health checks may fail due to insufficient GPU access, missing NGC or Kafka credentials, or incorrect Docker Compose profile configuration. The workflow provides error handling to diagnose these common deployment issues.