Inferra Deployment and Usage Guide
This guide takes an operator from a bare Kubernetes cluster to a validated, observable Inferra deployment. It targets Inferra 1.2.0.
Cluster-admin access to a Kubernetes cluster and a Cloudsmith entitlement token. Both are prerequisites for every page that follows — see Prerequisites.
What is in this guide
The step-by-step procedure for attaching the connector to a vLLM serving pod is in the companion vLLM Connector Guide. The vLLM Connector explains what that connector is, and Why the vLLM Pod Needs Host IPC covers the one deployment property it depends on.
Getting These Documents
The Inferra documents are published to Cloudsmith alongside the software, so you can always fetch the set that matches the build you installed. Each is published as a Markdown file; Word (.docx) copies are not published.
Cloudsmith serves them as raw packages: the download is the file itself, with nothing to unpack. There is no directory listing, so the URL must name the file exactly — and note that it carries the version twice, in two different forms: the PEP 440 form in the versions/ segment and the SemVer form in the filename. This is the same split the lcf-cli download in LCF-CLI Deployment uses.
Each download is a plain Markdown file, not an archive. Open it in any text editor or Markdown viewer.
The release notes carry the known issues for the release. They are updated between releases as issues are found or fixed, so re-fetch them if you are diagnosing something rather than relying on the copy you downloaded at install time. The current set is also published here: Release Notes.
Values to Substitute Before Running
vX.Y.Z/X.Y.Z— the Inferra release version and chart tag. This guide targets 1.2.0 (PEP 440: 1.2.0); use that unless Lightbits Support advises otherwise.<CLOUDSMITH_TOKEN>— your Cloudsmith read/entitlement token. Provided by Lightbits Support.