From f370813d7b9f6e80517efd7fe7ff249789a3e265 Mon Sep 17 00:00:00 2001 From: Ankit Anand Date: Sat, 29 Aug 2026 12:41:26 -0700 Subject: [PATCH] docs(self-managed): inventory optional ModelExpress artifacts in the manifest Signed-off-by: Ankit Anand --- docs/user/manifest.md | 2 ++ docs/version-catalog/main.yaml | 16 ++++++++++++++++ 2 files changed, 18 insertions(+) diff --git a/docs/user/manifest.md b/docs/user/manifest.md index f0a4aa0a1..38dad605d 100644 --- a/docs/user/manifest.md +++ b/docs/user/manifest.md @@ -211,6 +211,7 @@ The following tables list the complete artifact inventory. | `ebs-csi-driver` | `supported` | Optional | Provides Amazon EBS persistent volumes for EKS clusters. | `https://kubernetes-sigs.github.io/aws-ebs-csi-driver` | [Upstream](https://github.com/kubernetes-sigs/aws-ebs-csi-driver) | | `gpu-operator` | `supported` | Required | Manages NVIDIA GPU software on Kubernetes nodes. | `https://helm.ngc.nvidia.com/nvidia` | [Upstream](https://github.com/NVIDIA/gpu-operator) | | `helm-nvca-operator` | `1.12.7` | Required | Deploys the NVCA operator and compute-plane integration. | `https://helm.ngc.nvidia.com/nvidia/nvcf/helm-nvca-operator:1.12.7` | [GitHub](https://github.com/NVIDIA/nvcf/tree/main/deploy/helm/nvca-operator) | +| `modelexpress` | `supported` | Optional | Distributes model weights peer-to-peer between Dynamo workers to reduce scale-out cold starts. Installed separately from the compute-plane stack. | `https://helm.ngc.nvidia.com/nvidia/ai-dynamo` | [Upstream](https://github.com/ai-dynamo/modelexpress) | | `nvcf-container-cache` | `0.25.22` | Optional | Deploys container image caching on GPU cluster nodes. | `https://helm.ngc.nvidia.com/nvidia/nvcf/nvcf-container-cache:0.25.22` | [GitHub](https://github.com/NVIDIA/nvcf/tree/main/deploy/helm/container-cache) | ### Compute plane services and images @@ -220,6 +221,7 @@ The following tables list the complete artifact inventory. | `ess-agent` | `1.3.1` | Required | Injects encrypted application secrets into function workloads. | `nvcr.io/nvidia/nvcf/ess-agent:1.3.1` | [GitHub](https://github.com/NVIDIA/nvcf/tree/main/src/compute-plane-services/ess-agent) | | `gpu-operator-validator` | `supported` | Required | Validates GPU Operator components on GPU nodes. | `https://catalog.ngc.nvidia.com/orgs/nvidia/teams/cloud-native/containers/gpu-operator-validator` | [Upstream](https://github.com/NVIDIA/gpu-operator) | | `k8s-device-plugin` | `supported` | Required | Advertises NVIDIA GPU resources to Kubernetes. | `https://catalog.ngc.nvidia.com/orgs/nvidia/teams/k8s/containers/device-plugin` | [Upstream](https://github.com/NVIDIA/k8s-device-plugin) | +| `modelexpress-server` | `supported` | Optional | Serves model weights to Dynamo workers over NIXL RDMA transports. | `https://catalog.ngc.nvidia.com/orgs/nvidia/teams/ai-dynamo/containers/modelexpress-server` | [Upstream](https://github.com/ai-dynamo/modelexpress) | | `nvca` | `3.0.3` | Required | Registers GPU clusters and orchestrates deployments in-cluster. | `nvcr.io/nvidia/nvcf/nvca:3.0.3` | [GitHub](https://github.com/NVIDIA/nvcf/tree/main/src/compute-plane-services/nvca) | | `nvca-operator` | `3.0.3` | Required | Reconciles NVCA resources and compute-plane configuration. | `nvcr.io/nvidia/nvcf/nvca-operator:3.0.3` | | | `nvcf-container-cache` | `v1.1.36` | Optional | Caches container image layers on GPU cluster nodes. | `nvcr.io/nvidia/nvcf/nvcf-container-cache:v1.1.36` | [GitHub](https://github.com/NVIDIA/nvcf/tree/main/deploy/helm/container-cache) | diff --git a/docs/version-catalog/main.yaml b/docs/version-catalog/main.yaml index 5b25545be..fd9d03b5d 100644 --- a/docs/version-catalog/main.yaml +++ b/docs/version-catalog/main.yaml @@ -598,6 +598,14 @@ manifest: requirement: optional description: Provides SMB persistent volumes for supported deployments. upstream_url: https://github.com/kubernetes-csi/csi-driver-smb + - name: modelexpress + version: supported + distribution: https://helm.ngc.nvidia.com/nvidia/ai-dynamo + plane: compute + kind: chart + requirement: optional + description: Distributes model weights peer-to-peer between Dynamo workers to reduce scale-out cold starts. Installed separately from the compute-plane stack. + upstream_url: https://github.com/ai-dynamo/modelexpress - artifact_id: nvca plane: compute kind: service-image @@ -655,6 +663,14 @@ manifest: requirement: required description: Advertises NVIDIA GPU resources to Kubernetes. upstream_url: https://github.com/NVIDIA/k8s-device-plugin + - name: modelexpress-server + version: supported + distribution: https://catalog.ngc.nvidia.com/orgs/nvidia/teams/ai-dynamo/containers/modelexpress-server + plane: compute + kind: service-image + requirement: optional + description: Serves model weights to Dynamo workers over NIXL RDMA transports. + upstream_url: https://github.com/ai-dynamo/modelexpress - artifact_id: pylon plane: compute kind: service-image