Other[AIW] NVIDIA VSS 3.2.1 with RAG on A100 40GB
Deploy NVIDIA Video Search and Summarization (VSS) in Luna Labs, then ask questions about a video and a document. Check the answers against the original sources and inspect the tools that produced them. Retrieval-augmented generation (RAG) finds document passages that help answer a question. In this lab, Elasticsearch stores the document embeddings in your assigned RAG namespace. 90 minutes on a prepared assignment with cached model files. Service startup can extend the session. Intermediate. By the end of this lab, you will be able to: Check your prepared Luna session. Deploy RAG, Elasticsearch, and VSS. Index a document and retrieve its exact knowledge code with a source citation. Upload and play a ten-second video, then check two visible facts in the answer. Inspect the video and document tools used to produce the answers. Basic experience running terminal commands is helpful. Each task explains what to run and how to check the result. The facilitator prepares the shared cluster and uses a protected host credential to cache model files. Luna assigns two namespaces, , and a unique browser URL when you launch the lab. It delivers the workshop files and opens a terminal showing your settings. Cluster access is configured for you. The key needs NGC Catalog access, permission to download the required models, accepted model terms, and an expiration covering your session. Every learner runs Task 2 and pastes their key at a hidden terminal prompt before deploying applications. The prompt stays blank while you paste; press Enter to submit. Prepared caches do not bypass this step. Keep the key out of shell commands, chat, and screenshots. Launch your assigned Luna lab and wait for preparation to succeed. Open the. and start with. . Complete the ten tasks in order, using the terminal and browser inside the Luna desktop. Each terminal task explains its purpose, the main commands inside its script, and the expected result. Run the prepared task command, then check its result before continuing. If you close the terminal, open another one; the next command reloads your session settings. Run. to display your namespace and URL summary again. | Tasks | What you will accomplish | |---|---| | 1–2 | Check your session and add your own NGC key. | | 3–4 | Start RAG and check Elasticsearch, models, and APIs. | | 5–7 | Deploy VSS and connect its video and document tools. | | 8 | Index a document containing your session knowledge code. | | 9 | Check video upload, full playback, video answers, and document retrieval. | | 10 | Record your results and hand back the session. |. You have completed the lab when all Task 9 checks pass and you have recorded your results in Task 10. After the facilitator accepts the cleanup handoff, end the Luna session. The pipeline removes your assigned resources, including the namespace Secrets. After cleanup is confirmed, revoke your NGC key if you created it only for this workshop. If a command fails, your terminal stays open. Give the facilitator the task number and error message. Keep registry credentials out of chat messages and screenshots. Alejandro Casas, Senior Principal Product Marketing Manager, OCI. Anurag Kuppala, Solution Architect, NVIDIA. For questions about this lab, contact. . Adapted from the supplied VSS A100 40GB guides and the AIQ Luna provisioning pattern. The workshop uses NVIDIA VSS, NVIDIA RAG, NVIDIA NIM, and Elasticsearch on Oracle Kubernetes Engine. October 24, 2026: Made learner-owned NGC key entry mandatory in Task 2 before application deployment. October 23, 2026: Expanded all ten tasks with purpose, command explanations, and checkpoints; corrected readiness and collection-name checks. October 22, 2026: Added the Elasticsearch deployment profile with three GPUs per learner and a single YAML pipeline configuration. October 21, 2026: Simplified task commands, terminal recovery guidance, and learner prerequisites.