NVIDIA NIM 3.0.0

Use these standalone rules for NVIDIA NIM for Large Language Models 3.0.0 in standard or Dynamo mode, including its OpenAI-compatible frontend, router, vLLM workers, Dynamo Operator, model recipes, management APIs, caches, adapters, configuration, and observability.

Version, offering, and architecture boundary

• Deploy the NIM LLM image by digest together with the model, model revision, profile, GPU SKU, precision, worker arguments, and recipe that NVIDIA validated for that image; do not replace one element independently and still call the deployment validated.
• Take the NIM image tag and digest, operator chart, and recipe versions from the NGC catalog or the NIM release notes rather than memory, leaving a placeholder that fails the deployment when you can't check, and keep them in a values or lock file that an update bot or a scheduled review keeps current. The 3.0.0 in this prompt's title names the release these rules were checked against, not a version to install.
• Treat NIM LLM 2.0 and 3.0 as parallel release series; choose standard or Dynamo mode explicitly and do not interpret a higher series number as feature parity or an automatic migration instruction.
• Record that 3.0.0 is not NIM Certified and does not carry NIM Certified enterprise lifecycle, CVE SLA, compliance, or NVIDIA Enterprise support promises; select a qualifying 2.0 Certified image when those guarantees are mandatory.
• Inventory every component of the deployment and record each part's version so it can be checked against its own advisories: for 3.0.0 the release notes list vLLM 0.26.0, a