NemoClaw + Dynamo + Nemotron-3 Super FP8#215
Open
antonslutskyms wants to merge 16 commits into
Open
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adds/updates the Azure Kubernetes Service (AKS) workshop under nim-deploy/cloud-service-providers/azure/workshops/aks-nemoclaw. The workshop shows NVIDIA NemoClaw agents using NVIDIA Dynamo inference on-cluster, with nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-FP8 as the reference model.
What’s in this tree
README.md — Workshop goals, plain-language governable application surface vs inference plane, how they connect (Mermaid), Azure AKS tables (identity, secrets, policy, GPU, PVC, scaling, LB), economics vs metered APIs (with illustrative charts), disaggregated serving overview, KV routing caveat for this Nemotron disagg recipe, links to upstream Nemotron recipes and recipe README (PVC / model-cache / prerequisites), and install/ops for deploy_nemoclaw_k8s.sh.
deploy_nemoclaw_k8s.sh — Optional Dynamo apply + wait, NemoClaw source checkout, linux/amd64 image build, ACR push with login retry, apply nemoclaw-install manifests with ACR name substitution.
nemoclaw-base/ / nemoclaw-install/ — DinD + workspace pattern, NemoClaw install, endpoint wiring toward Dynamo frontend.
dynamo/ — Vendored Dynamo tree including default dynamo/dynamo/recipes/nemotron-3-super-fp8/sglang/disagg/deploy.yaml (other recipe paths documented: agg / TRT-LLM disagg, etc.).
How to test / review
README renders (tables + Mermaid) on your Git host.
Links to GitHub
ai-dynamo/dynamo
recipe paths resolve.
./deploy_nemoclaw_k8s.sh --help
matches documented flags and env vars.
(If code changed) Run deploy against a dev ACR + AKS with GPU recipe prerequisites satisfied.
Notes / follow-ups
Default inference manifest is SGLang disaggregated; DYNAMO_DEPLOY_MANIFEST can point at other upstream deploy.yaml layouts.
Nemotron KV-aware routing limitations for this disagg path are documented; align with upstream recipe README if behavior changes.
Related
Upstream recipes: https://github.com/ai-dynamo/dynamo/tree/main/recipes/nemotron-3-super-fp8
Recipe README (prereqs, PVC, quick start): https://github.com/ai-dynamo/dynamo/blob/main/recipes/nemotron-3-super-fp8/README.md