Choose a NIMService model variant

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/standalone/autoscaling/llm.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is d078fa423dbb9be6197e33568c0e222e8aa46039e18ee2f470901ba01b00a2fa
• objectSetSha256 is 03f7b8ce3ff95f8b7933fda5ab373c27469717e0f8670121f9d9551cebbbac63
• image is nvcr.io/nim/meta/llama-3.2-1b-instruct:1.12.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMCache spec.source.ngc.pullSecret and NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMCache spec.source.ngc.authSecret and NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.
• The sample enables a ServiceMonitor labelled release=kube-prometheus-stack, so the Prometheus Operator's ServiceMonitor API must exist on the destination. Declared. Not checked on any destination.
• The sample enables a HorizontalPodAutoscaler that reads the metric gpu_cache_usage_perc, so a metrics adapter on the destination has to serve it. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimcaches.apps.nvidia.com and nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMCache spec.source.ngc.pullSecret, NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret ngc-api-secret (NIMCache spec.source.ngc.authSecret, NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService reads its model from NIMCache meta-llama-3-2-1b-instruct, which this sample file also defines. NIMCache meta-llama-3-2-1b-instruct asks the operator to create a 50Gi ReadWriteOnce PersistentVolumeClaim with an empty storageClass, so the storage class is a destination choice. The listing gives the rest. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/standalone/autoscaling/multi-llm.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is 3c599dbaa3ea9ff7081db63134259f95174c52937b085249eeba3960015a711a
• objectSetSha256 is d0fbdfff38eaef948494be36233d6da35694d03091b16523ccac4d5d7b48af79
• image is nvcr.io/nim/nvidia/llm-nim:1.15.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMCache spec.source.hf.pullSecret and NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMCache spec.source.hf.authSecret. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.
• The sample enables a ServiceMonitor labelled release=kube-prometheus-stack, so the Prometheus Operator's ServiceMonitor API must exist on the destination. Declared. Not checked on any destination.
• The sample enables a HorizontalPodAutoscaler that reads the metric gpu_cache_usage_perc, so a metrics adapter on the destination has to serve it. Declared. Not checked on any destination.
• The sample fetches model weights for meta-llama/Llama-3.2-1B-Instruct from https://huggingface.co. Those weights carry their own licence, and this entry fetched none of them. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimcaches.apps.nvidia.com and nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMCache spec.source.hf.pullSecret, NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret hf-api-secret (NIMCache spec.source.hf.authSecret); ngc-api-secret (NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService reads its model from NIMCache nim-cache-multi-llm, which this sample file also defines. NIMCache nim-cache-multi-llm asks the operator to create a 50Gi ReadWriteOnce PersistentVolumeClaim with an empty storageClass, so the storage class is a destination choice. The listing gives the rest. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/standalone/basic/llm.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is 4218536519a5fdea32d12c719b7e37a93c0d5f4b223c854599d984caccc7a1a6
• objectSetSha256 is b5943236fcd709b567a1f932a847c169a7bc6e9a40dac141e6fac4cb9a1fb797
• image is nvcr.io/nim/meta/llama-3.2-1b-instruct:1.12.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMCache spec.source.ngc.pullSecret and NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMCache spec.source.ngc.authSecret and NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimcaches.apps.nvidia.com and nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMCache spec.source.ngc.pullSecret, NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret ngc-api-secret (NIMCache spec.source.ngc.authSecret, NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService reads its model from NIMCache meta-llama-3-2-1b-instruct, which this sample file also defines. NIMCache meta-llama-3-2-1b-instruct asks the operator to create a 50Gi ReadWriteOnce PersistentVolumeClaim with an empty storageClass, so the storage class is a destination choice. The listing gives the rest. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/standalone/basic/multi-llm.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is ce7c2a54d6cf71c90ca9947697ad3565289b6811660b33ca0722a8dd1ba31e7c
• objectSetSha256 is 2763aee71a2ea72ece365255db4e3851eed92867d1c1e105c4a694ba7bf11989
• image is nvcr.io/nim/nvidia/llm-nim:1.15.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMCache spec.source.hf.pullSecret and NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMCache spec.source.hf.authSecret. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.
• The sample fetches model weights for meta-llama/Llama-3.2-1B-Instruct from https://huggingface.co. Those weights carry their own licence, and this entry fetched none of them. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimcaches.apps.nvidia.com and nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMCache spec.source.hf.pullSecret, NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret hf-api-secret (NIMCache spec.source.hf.authSecret); ngc-api-secret (NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService reads its model from NIMCache nim-cache-multi-llm, which this sample file also defines. NIMCache nim-cache-multi-llm asks the operator to create a 50Gi ReadWriteOnce PersistentVolumeClaim with an empty storageClass, so the storage class is a destination choice. The listing gives the rest. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/standalone/basic/model-free.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is 2b56ea5fbd01729d15aedc32253c6a997c89465a54fc33ce36fd0f1f24b519be
• objectSetSha256 is 0770f9dfbc1003416eb1e66523a9ee2a810e3692cde284249ea1049f8d9643ed
• image is nvcr.io/nim/nvidia/model-free-nim:3.0.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMCache spec.source.hf.pullSecret and NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMCache spec.source.hf.authSecret and NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.
• The sample fetches model weights for meta-llama/Llama-3.2-1B-Instruct from https://huggingface.co. Those weights carry their own licence, and this entry fetched none of them. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimcaches.apps.nvidia.com and nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMCache spec.source.hf.pullSecret, NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret hf-api-secret (NIMCache spec.source.hf.authSecret, NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService reads its model from NIMCache nim-cache-model-free, which this sample file also defines. NIMCache nim-cache-model-free asks the operator to create a 50Gi ReadWriteOnce PersistentVolumeClaim with an empty storageClass, so the storage class is a destination choice. The listing gives the rest. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/standalone/confidential-computing/llm-kata-sandbox.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is b47f6df0183fb5ceefea7f9751adbe4952f3fc64287a1627050c023825a9f462
• objectSetSha256 is 4582a67b8e5bcff226b91319a744823e84308d81f05fa5827b3e225641012354
• image is nvcr.io/nim/meta/llama-3.2-1b-instruct:1.12.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod as nvidia.com/pgpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.
• The sample sets runtimeClassName to kata-qemu-nvidia-gpu, so that RuntimeClass must exist on the destination. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret ngc-api-secret (NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService uses an emptyDir volume limited to 10Gi, so the model is fetched again whenever the pod is replaced. Nothing here checked that a destination can provide it. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/advanced/epp/nimservice.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is eb3952f4209f4b63a9cce3e232b28c01016c55fe28e3128ae0cbda3559382668
• objectSetSha256 is b8940509a487ac76e3ed2e4f5f619e28d0ed51814f6b187ce8a7fde0202b06a6
• image is nvcr.io/nvstaging/nim-internal/vllm-oss-nim-llm-4-devel-llama3.1-8b-instruct:2.0.0-rc.20260203175658-38778af4c5e1fc45

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod as nvidia.com/gpu, with 2 replicas. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• The sample sets no namespace, so the destination chooses one and it must exist before apply. Declared. Not checked on any destination.
• spec.expose.router.gateway names the Gateway inference-gateway in namespace nim-service and asks for HTTPRoute objects. The Gateway API and that Gateway must exist on the destination. Declared. Not checked on any destination.
• spec.expose.router.gateway.backendRef names an InferencePool in the inference.networking.k8s.io API group, so that API must be installed on the destination. Declared. Not checked on any destination.
• spec.expose.router.hostDomainName is demo.nvidia.example.com, an example domain. A destination supplies its own. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret ngc-api-secret (NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService reads its model from NIMCache meta-llama3-8b-instruct, which this sample file does not define. Nothing here checked that a destination can provide it. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/advanced/dra/auto-creation/example0/llm.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is 40a21f78f2ddbccfe2d3bf67c961d032bd1b0d9cd74ddc523eaa9f718fd658b0
• objectSetSha256 is 86f43ab875e8756e03fef5d9fe148b8bf79eb6634e3227cd7563718d3fb5dc26
• image is nvcr.io/nim/meta/llama-3.2-1b-instruct:1.12.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMCache spec.source.ngc.pullSecret and NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMCache spec.source.ngc.authSecret and NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod through a Dynamic Resource Allocation claim, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.
• The sample claims its GPU through Dynamic Resource Allocation with device class gpu.nvidia.com. The destination needs the resource.k8s.io API and a driver that publishes that class. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimcaches.apps.nvidia.com and nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMCache spec.source.ngc.pullSecret, NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret ngc-api-secret (NIMCache spec.source.ngc.authSecret, NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService reads its model from NIMCache meta-llama3-2-1b-instruct, which this sample file also defines. NIMCache meta-llama3-2-1b-instruct asks the operator to create a 50Gi ReadWriteOnce PersistentVolumeClaim with an empty storageClass, so the storage class is a destination choice. The listing gives the rest. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/advanced/dra/auto-creation/example1/llm.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is 6b79a2643dcafb117cc0dc9a1a4c02b310d2a9c14e900bfedb9b6b1899fffc51
• objectSetSha256 is d2e5aaa047a10663926a0d8fe3d83add1e2170d16c87c6305da4db2e6ea7154f
• image is nvcr.io/nim/meta/llama-3.2-1b-instruct:1.12.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMCache spec.source.ngc.pullSecret and NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMCache spec.source.ngc.authSecret and NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod through a Dynamic Resource Allocation claim, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.
• The sample claims its GPU through Dynamic Resource Allocation with device class gpu.nvidia.com. The destination needs the resource.k8s.io API and a driver that publishes that class. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimcaches.apps.nvidia.com and nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMCache spec.source.ngc.pullSecret, NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret ngc-api-secret (NIMCache spec.source.ngc.authSecret, NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService reads its model from NIMCache meta-llama3-2-1b-instruct, which this sample file also defines. NIMCache meta-llama3-2-1b-instruct asks the operator to create a 50Gi ReadWriteOnce PersistentVolumeClaim with an empty storageClass, so the storage class is a destination choice. The listing gives the rest. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/advanced/dra/auto-creation/example2/llm.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is 18f5143972d25269600890a3dc0051391709bdd110fe5b3c2d87d3fb5106d3f0
• objectSetSha256 is 4f4bf1f156bd8621fd4706ed4b23e565dbeb229fdd05fe6034d1ad94d9075ab5
• image is nvcr.io/nim/meta/llama-3.2-1b-instruct:1.12.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMCache spec.source.ngc.pullSecret and NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMCache spec.source.ngc.authSecret and NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod through a Dynamic Resource Allocation claim, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.
• The sample claims its GPU through Dynamic Resource Allocation with device class gpu.nvidia.com. The destination needs the resource.k8s.io API and a driver that publishes that class. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimcaches.apps.nvidia.com and nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMCache spec.source.ngc.pullSecret, NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret ngc-api-secret (NIMCache spec.source.ngc.authSecret, NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService reads its model from NIMCache meta-llama3-2-1b-instruct, which this sample file also defines. NIMCache meta-llama3-2-1b-instruct asks the operator to create a 50Gi ReadWriteOnce PersistentVolumeClaim with an empty storageClass, so the storage class is a destination choice. The listing gives the rest. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/standalone/full/llm.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is 48834814da4594e44f899683e1f5fb3984c42b1514d11f6c1a3dd1e4de46dc94
• objectSetSha256 is b56d4ecd50613355fb5cd6a75b6b20b5451323bc810fac0f3a1a917ae1e8d436
• image is nvcr.io/nim/meta/llama-3.2-1b-instruct:1.12.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMCache spec.source.ngc.pullSecret and NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMCache spec.source.ngc.authSecret and NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.
• spec.expose.ingress names the ingress class nginx, so an ingress controller serving that class must run on the destination. Declared. Not checked on any destination.
• spec.expose.ingress routes the host demo.nvidia.example.com, an example domain. A destination supplies its own. Declared. Not checked on any destination.
• The sample enables a ServiceMonitor labelled release=kube-prometheus-stack, so the Prometheus Operator's ServiceMonitor API must exist on the destination. Declared. Not checked on any destination.
• The sample enables a HorizontalPodAutoscaler that reads the metric gpu_cache_usage_perc, so a metrics adapter on the destination has to serve it. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimcaches.apps.nvidia.com and nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMCache spec.source.ngc.pullSecret, NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret ngc-api-secret (NIMCache spec.source.ngc.authSecret, NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService reads its model from NIMCache meta-llama-3-2-1b-instruct, which this sample file also defines. NIMCache meta-llama-3-2-1b-instruct asks the operator to create a 50Gi ReadWriteOnce PersistentVolumeClaim with an empty storageClass, so the storage class is a destination choice. The listing gives the rest. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/standalone/full/multi-llm.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is cf2fcef92859da8db83cd4f529d7a44c8b13afcb43af9ad221e11d23a929ab21
• objectSetSha256 is 8747029d5cbb287735564bc069b90f24333a12141e5d4bb2a7fe65203f5c3555
• image is nvcr.io/nim/nvidia/llm-nim:1.15.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMCache spec.source.hf.pullSecret and NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMCache spec.source.hf.authSecret. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.
• spec.expose.ingress names the ingress class nginx, so an ingress controller serving that class must run on the destination. Declared. Not checked on any destination.
• spec.expose.ingress routes the host demo.nvidia.example.com, an example domain. A destination supplies its own. Declared. Not checked on any destination.
• The sample enables a ServiceMonitor labelled release=kube-prometheus-stack, so the Prometheus Operator's ServiceMonitor API must exist on the destination. Declared. Not checked on any destination.
• The sample enables a HorizontalPodAutoscaler that reads the metric gpu_cache_usage_perc, so a metrics adapter on the destination has to serve it. Declared. Not checked on any destination.
• The sample fetches model weights for meta-llama/Llama-3.2-1B-Instruct from https://huggingface.co. Those weights carry their own licence, and this entry fetched none of them. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimcaches.apps.nvidia.com and nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMCache spec.source.hf.pullSecret, NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret hf-api-secret (NIMCache spec.source.hf.authSecret); ngc-api-secret (NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService reads its model from NIMCache nim-cache-multi-llm, which this sample file also defines. NIMCache nim-cache-multi-llm asks the operator to create a 50Gi ReadWriteOnce PersistentVolumeClaim with an empty storageClass, so the storage class is a destination choice. The listing gives the rest. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/standalone/routers/grpcroutes/llm.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is daa8e038c7e4a6cb6c540094e607da9066fd23356a72c2e1d9d3a57302d281a9
• objectSetSha256 is 521b278642e743b0c7233c8da8cc9238397ffae08ebbbc3632ad4e5fea0706c7
• image is nvcr.io/nim/nvidia/riva-tts:1.3.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMCache spec.source.ngc.pullSecret and NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMCache spec.source.ngc.authSecret and NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.
• spec.expose.router.gateway names the Gateway istio-gateway in namespace nemo and asks for GRPCRoute objects. The Gateway API and that Gateway must exist on the destination. Declared. Not checked on any destination.
• spec.expose.router.hostDomainName is demo.nvidia.example.com, an example domain. A destination supplies its own. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimcaches.apps.nvidia.com and nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMCache spec.source.ngc.pullSecret, NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret ngc-api-secret (NIMCache spec.source.ngc.authSecret, NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService reads its model from NIMCache riva-tts, which this sample file also defines. NIMCache riva-tts asks the operator to create a 50Gi ReadWriteOnce PersistentVolumeClaim with an empty storageClass, so the storage class is a destination choice. The listing gives the rest. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/standalone/routers/httproutes/llm.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is cfcb9c6cd91e8c90f4146de2f642c41890cff6075f0b6619b60333a0bc0539b9
• objectSetSha256 is 960b72663a10608beba584093e615248726e1e36ddce3cd880ca27a1e514c85b
• image is nvcr.io/nim/meta/llama-3.2-1b-instruct:1.12.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMCache spec.source.ngc.pullSecret and NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMCache spec.source.ngc.authSecret and NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.
• spec.expose.router.gateway names the Gateway istio-gateway in namespace nim-service and asks for HTTPRoute objects. The Gateway API and that Gateway must exist on the destination. Declared. Not checked on any destination.
• spec.expose.router.hostDomainName is demo.nvidia.example.com, an example domain. A destination supplies its own. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimcaches.apps.nvidia.com and nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMCache spec.source.ngc.pullSecret, NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret ngc-api-secret (NIMCache spec.source.ngc.authSecret, NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService reads its model from NIMCache meta-llama-3-2-1b-instruct, which this sample file also defines. NIMCache meta-llama-3-2-1b-instruct asks the operator to create a 50Gi ReadWriteOnce PersistentVolumeClaim with an empty storageClass, so the storage class is a destination choice. The listing gives the rest. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/standalone/routers/httproutes/multi-llm.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is 52508df4e4df05c023b399449dcc9fdbab69c72b64b2c2c8afbc27f7be69011d
• objectSetSha256 is 34ddfe463ffb07c596dfe62566f4942ce4daf946400d645b97aa8e6b8071bc74
• image is nvcr.io/nim/nvidia/llm-nim:1.15.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMCache spec.source.hf.pullSecret and NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMCache spec.source.hf.authSecret. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.
• spec.expose.router.gateway names the Gateway istio-gateway in namespace nim-service and asks for HTTPRoute objects. The Gateway API and that Gateway must exist on the destination. Declared. Not checked on any destination.
• spec.expose.router.hostDomainName is demo.nvidia.example.com, an example domain. A destination supplies its own. Declared. Not checked on any destination.
• The sample fetches model weights for meta-llama/Llama-3.2-1B-Instruct from https://huggingface.co. Those weights carry their own licence, and this entry fetched none of them. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimcaches.apps.nvidia.com and nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMCache spec.source.hf.pullSecret, NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret hf-api-secret (NIMCache spec.source.hf.authSecret); ngc-api-secret (NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService reads its model from NIMCache nim-cache-multi-llm, which this sample file also defines. NIMCache nim-cache-multi-llm asks the operator to create a 50Gi ReadWriteOnce PersistentVolumeClaim with an empty storageClass, so the storage class is a destination choice. The listing gives the rest. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/standalone/routers/ingress/llm.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is 47f4da1567742044a469ef5a2ff398f0822b24d246307990640eee3e668e2beb
• objectSetSha256 is c41323262ec969726b69d2e4d930ee0c5007e8a969434d34781c3fd9cc1451f8
• image is nvcr.io/nim/meta/llama-3.2-1b-instruct:1.12.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >3.1.0 must be installed first. The sample sets NIMService spec.router, which the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds do not declare. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMCache spec.source.ngc.pullSecret and NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMCache spec.source.ngc.authSecret and NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.
• spec.router.ingress names the ingress class nginx, so an ingress controller serving that class must run on the destination. Declared. Not checked on any destination.
• spec.router.hostDomainName is demo.example.nvidia.com, an example domain. A destination supplies its own. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >3.1.0 and wait for nimcaches.apps.nvidia.com and nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMCache spec.source.ngc.pullSecret, NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret ngc-api-secret (NIMCache spec.source.ngc.authSecret, NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService reads its model from NIMCache meta-llama-3-2-1b-instruct, which this sample file also defines. NIMCache meta-llama-3-2-1b-instruct asks the operator to create a 50Gi ReadWriteOnce PersistentVolumeClaim with an empty storageClass, so the storage class is a destination choice. The listing gives the rest. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/standalone/routers/ingress/multi-llm.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is baa21918b9ff3d9d54713c6dfa8b8207b8189840b9c28a24247da8cae762a905
• objectSetSha256 is 7c83ab4df35cb94d39bed54dcce7333e3ded8573db12485bb8897b128ec79dc7
• image is nvcr.io/nim/nvidia/llm-nim:1.15.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMCache spec.source.hf.pullSecret and NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMCache spec.source.hf.authSecret. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.
• spec.expose.router.ingress names the ingress class nginx, so an ingress controller serving that class must run on the destination. Declared. Not checked on any destination.
• spec.expose.router.hostDomainName is demo.nvidia.example.com, an example domain. A destination supplies its own. Declared. Not checked on any destination.
• The sample fetches model weights for meta-llama/Llama-3.2-1B-Instruct from https://huggingface.co. Those weights carry their own licence, and this entry fetched none of them. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimcaches.apps.nvidia.com and nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMCache spec.source.hf.pullSecret, NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret hf-api-secret (NIMCache spec.source.hf.authSecret); ngc-api-secret (NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService reads its model from NIMCache nim-cache-multi-llm, which this sample file also defines. NIMCache nim-cache-multi-llm asks the operator to create a 50Gi ReadWriteOnce PersistentVolumeClaim with an empty storageClass, so the storage class is a destination choice. The listing gives the rest. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/kserve/knative/llm.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is e260f8b3941ea74d3611923473d9c76da0d073b8faffa83fd8c04a381778d90b
• objectSetSha256 is db534c939adf26e2dc9a7e927e4c4b2f19f9cac1af91bf9178282156bf786670
• image is nvcr.io/nim/meta/llama-3.2-1b-instruct:1.12.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMCache spec.source.ngc.pullSecret and NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMCache spec.source.ngc.authSecret and NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.
• The sample sets spec.inferencePlatform to kserve, so that platform must be installed on the destination before the operator can serve this NIMService. Declared. Not checked on any destination.
• The sample carries autoscaling.knative.dev annotations, so Knative Serving must be installed on the destination. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimcaches.apps.nvidia.com and nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMCache spec.source.ngc.pullSecret, NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret ngc-api-secret (NIMCache spec.source.ngc.authSecret, NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService reads its model from NIMCache meta-llama-3-2-1b-instruct, which this sample file also defines. NIMCache meta-llama-3-2-1b-instruct asks the operator to create a 50Gi ReadWriteOnce PersistentVolumeClaim with an empty storageClass, so the storage class is a destination choice. The listing gives the rest. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/kserve/knative/multi-llm.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is 6555417d54f00bbb0335fb4fb12cf01bbfb9682066347adbcfaad9ec51786c82
• objectSetSha256 is b5020b77339bf0181254255a04927e5ce4775a345fc060d48d91cb070ae278fd
• image is nvcr.io/nim/nvidia/llm-nim:1.15.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMCache spec.source.hf.pullSecret and NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMCache spec.source.hf.authSecret. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.
• The sample sets spec.inferencePlatform to kserve, so that platform must be installed on the destination before the operator can serve this NIMService. Declared. Not checked on any destination.
• The sample carries autoscaling.knative.dev annotations, so Knative Serving must be installed on the destination. Declared. Not checked on any destination.
• The sample fetches model weights for nvidia/Llama-3.1-Nemotron-Nano-8B-v1 from https://huggingface.co. Those weights carry their own licence, and this entry fetched none of them. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimcaches.apps.nvidia.com and nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMCache spec.source.hf.pullSecret, NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret hf-api-secret (NIMCache spec.source.hf.authSecret); ngc-api-secret (NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService reads its model from NIMCache nim-cache-multi-llm, which this sample file also defines. NIMCache nim-cache-multi-llm asks the operator to create a 50Gi ReadWriteOnce PersistentVolumeClaim with an empty storageClass, so the storage class is a destination choice. The listing gives the rest. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/standalone/lora/llm.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is a4afe2337f856a22fb4837b9b24931044f60a64c0d89fd5f77776eac92cd170a
• objectSetSha256 is 05a65e0d30de4763bacbcb1a4e961ff1c6711955c30698e43db3d3481c654a4a
• image is nvcr.io/nim/meta/llama-3.2-1b-instruct:1.12.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret ngc-api-secret (NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. NIMService meta-llama-3-2-1b-instruct expects a PersistentVolumeClaim named meta-llama-3-2-1b-instruct with an empty storageClass, so the storage class is a destination choice. Nothing here checked that a destination can provide it. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/standalone/lora/multi-llm.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is 73d22658c234f33b57e8e5c016d70d6578b825920234cb2d942a47e0d9e9549e
• objectSetSha256 is b8b3a49f19e059947ea276ba0b9df4b691bf2df1010fb8d631de57d9c22a78e6
• image is nvcr.io/nim/nvidia/llm-nim:1.15.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret ngc-api-secret (NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. NIMService meta-llama-3-2-1b-instruct expects a PersistentVolumeClaim named nim-cache-multi-llm with an empty storageClass, so the storage class is a destination choice. Nothing here checked that a destination can provide it. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/advanced/dra/manual/llm.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is 3d351f9fe02750994c647e79e5405b277bee11f75bb1dfa0e9e589598a8a0ac5
• objectSetSha256 is bdf8717cd1dbe0e3f8f469bd5075fc7312fd6776e698d233e3c59a5b966ad317
• image is nvcr.io/nim/meta/llama-3.2-1b-instruct:1.12.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMCache spec.source.ngc.pullSecret and NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMCache spec.source.ngc.authSecret and NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod through a Dynamic Resource Allocation claim, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.
• The sample claims its GPU through Dynamic Resource Allocation with device class gpu.nvidia.com. The destination needs the resource.k8s.io API and a driver that publishes that class. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimcaches.apps.nvidia.com and nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMCache spec.source.ngc.pullSecret, NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret ngc-api-secret (NIMCache spec.source.ngc.authSecret, NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService reads its model from NIMCache meta-llama3-2-1b-instruct, which this sample file also defines. NIMCache meta-llama3-2-1b-instruct asks the operator to create a 50Gi ReadWriteOnce PersistentVolumeClaim with an empty storageClass, so the storage class is a destination choice. The listing gives the rest. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/advanced/multi-node/multi-node-nimservice.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is 713ddd5714069af7e5618d94650409427010819ae541590c091e15b182172607
• objectSetSha256 is 62a86e115f6282d79fc2f58b0b3bae264a679e9e3fce49e714959cbd9e644218
• image is nvcr.io/nim/deepseek-ai/deepseek-r1:1.7.3

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMCache spec.source.ngc.pullSecret and NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMCache spec.source.ngc.authSecret and NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 8 GPUs per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.
• The sample sets spec.multiNode (mpi), so one model spans several pods. The destination needs whatever the operator uses to group those pods, and this entry has not established what that is. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimcaches.apps.nvidia.com and nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMCache spec.source.ngc.pullSecret, NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret ngc-api-secret (NIMCache spec.source.ngc.authSecret, NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService reads its model from NIMCache deepseek-r1-nimcache, which this sample file also defines. NIMCache deepseek-r1-nimcache asks the operator to create a 100Gi ReadWriteMany PersistentVolumeClaim with an empty storageClass, so the storage class is a destination choice. The listing gives the rest. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/advanced/multi-node/multi-node-nimservice-rdma.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is f0672c0b775a61b57f937e73f57e55f1f10e70cc047730733e49710d6707aedc
• objectSetSha256 is fc9589debd11f39120dfbae346a6983e546dc42950e5d0f435360c5ee9b900a2
• image is nvcr.io/nim/deepseek-ai/deepseek-r1:1.7.3

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMCache spec.source.ngc.pullSecret and NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMCache spec.source.ngc.authSecret and NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 8 GPUs per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.
• The NIMService requests the extended resource rdma/rdma_shared_device_a=1, so a device plugin on the destination has to advertise it. Declared. Not checked on any destination.
• The sample attaches the pods to the secondary network example-ipoibnetwork, so a network attachment of that name must exist on the destination. Declared. Not checked on any destination.
• The sample sets spec.multiNode (mpi), so one model spans several pods. The destination needs whatever the operator uses to group those pods, and this entry has not established what that is. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimcaches.apps.nvidia.com and nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMCache spec.source.ngc.pullSecret, NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret ngc-api-secret (NIMCache spec.source.ngc.authSecret, NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService reads its model from NIMCache deepseek-r1-nimcache, which this sample file also defines. NIMCache deepseek-r1-nimcache asks the operator to create a 100Gi ReadWriteMany PersistentVolumeClaim with an empty storageClass, so the storage class is a destination choice. The listing gives the rest. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/advanced/multi-node/multi-node-nimservice-ray.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is 75f383e42735d9f5514cc6c5f3c05f74600953ce5944febd4167582f865db44d
• objectSetSha256 is 0171b9583cea00bade0db671d012a5702e11f91154fe01efaf6bd69079caf0c9
• image is nvcr.io/nim/meta/llama-3.1-8b-instruct:2.0.10

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >3.1.0 must be installed first. The sample sets NIMService spec.multiNode.ray, which the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds do not declare. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMCache spec.source.ngc.pullSecret and NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMCache spec.source.ngc.authSecret and NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 2 GPUs per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.
• The sample sets spec.multiNode (ray), so one model spans several pods. The destination needs whatever the operator uses to group those pods, and this entry has not established what that is. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >3.1.0 and wait for nimcaches.apps.nvidia.com and nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMCache spec.source.ngc.pullSecret, NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret ngc-api-secret (NIMCache spec.source.ngc.authSecret, NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService reads its model from NIMCache llama-3-1-8b-instruct-multinode, which this sample file also defines. NIMCache llama-3-1-8b-instruct-multinode asks the operator to create a 100Gi ReadWriteMany PersistentVolumeClaim with an empty storageClass, so the storage class is a destination choice. The listing gives the rest. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/standalone/nim-3/nemotron-3-super.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is 20a835700d82d1fc5c61cbec9b83248d51e68ab104c8bbdfa25fed0f7e3acbf8
• objectSetSha256 is c72d8fb61e0f8d4df55ce4e0d1f08f51f580561539af2aad16f1035c0e52221c
• image is nvcr.io/nim/nvidia/nemotron-3-super-120b-a12b:3.0.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMCache spec.source.ngc.pullSecret and NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMCache spec.source.ngc.authSecret and NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 4 GPUs per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimcaches.apps.nvidia.com and nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMCache spec.source.ngc.pullSecret, NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret ngc-api-secret (NIMCache spec.source.ngc.authSecret, NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService reads its model from NIMCache nemotron-3-super-120b-a12b, which this sample file also defines. NIMCache nemotron-3-super-120b-a12b asks the operator to create a 250Gi ReadWriteOnce PersistentVolumeClaim with an empty storageClass, so the storage class is a destination choice. The listing gives the rest. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/standalone/no-precaching/llm-emptydir.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is 8b39f15563795b45a794eefc381233c2008899bf984e23b656653169215a004d
• objectSetSha256 is 342ff6df188236559b665aa632f6395f604eee910b3586e7ed1c9c3bfc246124
• image is nvcr.io/nim/meta/llama-3.2-1b-instruct:1.12.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret ngc-api-secret (NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService uses an emptyDir volume limited to 10Gi, so the model is fetched again whenever the pod is replaced. Nothing here checked that a destination can provide it. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/standalone/no-precaching/llm-hostpath.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is da0727a0adbe8f036fca448fee699e2d6b25a48398b940ee7b66bda955e0ec1e
• objectSetSha256 is 4095730535defb640cbc93e67a813d7706c625638995a60dbd0f9b0ac9ac0a8a
• image is nvcr.io/nim/meta/llama-3.2-1b-instruct:1.12.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret ngc-api-secret (NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService mounts the node path /mnt/models/llama-3.2-1b-instruct, so every node that can run it must hold that path. Nothing here checked that a destination can provide it. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/standalone/no-precaching/llm.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is 0fe325030835acdf64a4ffe39a62c719ece56938e59a1ecb936d5a611cfa4054
• objectSetSha256 is b990826165eb254a9e6061d41993883d42bfb874fec899020e3a898600ea57c5
• image is nvcr.io/nim/meta/llama-3.2-1b-instruct:1.12.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret ngc-api-secret (NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. NIMService meta-llama-3-2-1b-instruct asks the operator to create a 10Gi ReadWriteOnce PersistentVolumeClaim with an empty storageClass, so the storage class is a destination choice. Nothing here checked that a destination can provide it. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/standalone/no-precaching/multi-llm-hf.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is 256b7baa7d532fd5a06beeeb0df2e69b847a7ac73b06dab391eabbb65e1c8dae
• objectSetSha256 is 45d247854cbd173d407cc12b2ac50788d0014be73d80d67926e9ce0979fd198d
• image is nvcr.io/nim/nvidia/llm-nim:1.15.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.
• The sample fetches model weights for meta-llama/Llama-3.2-1B-Instruct from Hugging Face. Those weights carry their own licence, and this entry fetched none of them. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret hf-secret (NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. NIMService meta-llama-3-2-1b-instruct asks the operator to create a 10Gi ReadWriteOnce PersistentVolumeClaim with an empty storageClass, so the storage class is a destination choice. Nothing here checked that a destination can provide it. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/standalone/no-precaching/multi-llm-ngc.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is 2de96273782232491a7cc8ad1feefaa03b9b5bab59eb2386455b869f1f2bcafa
• objectSetSha256 is e4d032885ab3d2e87043560d8f3f6f58f2c02c6353ff1bf4cafdb1f938e6c6a0
• image is nvcr.io/nim/nvidia/llm-nim:1.15.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret ngc-api-secret (NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. NIMService meta-llama-3-2-1b-instruct asks the operator to create a 10Gi ReadWriteOnce PersistentVolumeClaim with an empty storageClass, so the storage class is a destination choice. Nothing here checked that a destination can provide it. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/standalone/no-precaching/model-free-hf.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is c2d3c95da50e4931b40d614e30e6750a171c4a3072a8d65995b940dbd2a3fba9
• objectSetSha256 is 1f3a2cacc484ae7e67bf8206ded6a4ba24f7aaa548c796622e3e2864b60fab2b
• image is nvcr.io/nim/nvidia/model-free-nim:3.0.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.
• The sample fetches model weights for meta-llama/Llama-3.2-1B-Instruct from Hugging Face. Those weights carry their own licence, and this entry fetched none of them. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret hf-secret (NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. NIMService model-free-llama-3-2-1b-instruct asks the operator to create a 50Gi ReadWriteOnce PersistentVolumeClaim with an empty storageClass, so the storage class is a destination choice. Nothing here checked that a destination can provide it. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/standalone/retriever/embed.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is c7058d52175464fba77c58613365c7c163034d9cd396ae3b33c87fe5ebc0fabe
• objectSetSha256 is 1aafdea7ffa3795ddbeb645b82d0f526af40645390f78d2b9d415e29d5ec364a
• image is nvcr.io/nim/nvidia/llama-nemotron-embed-vl-1b-v2:2.3.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMCache spec.source.ngc.pullSecret and NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMCache spec.source.ngc.authSecret and NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimcaches.apps.nvidia.com and nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMCache spec.source.ngc.pullSecret, NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret hf-api-secret (NIMCache spec.source.ngc.authSecret, NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService reads its model from NIMCache llama-nemotron-embed-vl-1b-v2, which this sample file also defines. NIMCache llama-nemotron-embed-vl-1b-v2 asks the operator to create a 50Gi ReadWriteOnce PersistentVolumeClaim with an empty storageClass, so the storage class is a destination choice. The listing gives the rest. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/standalone/scheduling/llm.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is d8ce2dbe3d62044a657f18613cdf8d39e346e20b4abc4a2444238736c65aacdb
• objectSetSha256 is d9b5b69bce17711febcb3531ed5f642e12c0e61043cf60b0074a4e1636bd3f6f
• image is nvcr.io/nim/meta/llama-3.2-1b-instruct:1.12.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >3.1.0 must be installed first. The sample sets NIMService spec.priorityClassName, which the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds do not declare. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.
• The sample sets priorityClassName to gpu-critical. The sample ships no PriorityClass, so one of that name must exist on the destination. Declared. Not checked on any destination.
• The sample schedules only onto nodes where nvidia.com/gpu.present In true, so the destination's nodes must carry that label. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >3.1.0 and wait for nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret ngc-api-secret (NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService uses an emptyDir volume limited to 10Gi, so the model is fetched again whenever the pod is replaced. Nothing here checked that a destination can provide it. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/standalone/sidecars/llm.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is a283dedc018fd5b19ad075fa1390204e6e61de22a5d22d132538305dfa43b634
• objectSetSha256 is 4a865b472acd71864c994a9ba280a2f4713f47f3274cfbb62f0b1e8f0b5eec3c
• image is nvcr.io/nim/meta/llama-3.2-1b-instruct:1.12.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret ngc-api-secret (NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService uses an emptyDir volume limited to 10Gi, so the model is fetched again whenever the pod is replaced. Nothing here checked that a destination can provide it. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/kserve/standard/llm.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is 92af1ce520b39464bf7cc4da855ea2e8163203ddfb3f5656700413e729ccac47
• objectSetSha256 is 24899d2cfed44bdbbeb4d23800b381fbf75150649946de3b2da9d81d43864359
• image is nvcr.io/nim/meta/llama-3.2-1b-instruct:1.12.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMCache spec.source.ngc.pullSecret and NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMCache spec.source.ngc.authSecret and NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.
• The sample sets spec.inferencePlatform to kserve, so that platform must be installed on the destination before the operator can serve this NIMService. Declared. Not checked on any destination.
• The sample enables a HorizontalPodAutoscaler on resource metrics, so the destination needs a metrics server. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimcaches.apps.nvidia.com and nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMCache spec.source.ngc.pullSecret, NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret ngc-api-secret (NIMCache spec.source.ngc.authSecret, NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService reads its model from NIMCache meta-llama-3-2-1b-instruct, which this sample file also defines. NIMCache meta-llama-3-2-1b-instruct asks the operator to create a 50Gi ReadWriteOnce PersistentVolumeClaim with an empty storageClass, so the storage class is a destination choice. The listing gives the rest. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/kserve/standard/model-free.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is adec6f4cb47e17dba45b40f177c7b8c206bd59607baed41bd5c7df5de6a38904
• objectSetSha256 is 41e2cd864f8bd9bf13dfa1acd1a201342561144b06e52d5d9050042174858609
• image is nvcr.io/nim/nvidia/model-free-nim:3.0.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMCache spec.source.hf.pullSecret and NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMCache spec.source.hf.authSecret and NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.
• The sample sets spec.inferencePlatform to kserve, so that platform must be installed on the destination before the operator can serve this NIMService. Declared. Not checked on any destination.
• The sample fetches model weights for meta-llama/Llama-3.2-1B-Instruct from https://huggingface.co. Those weights carry their own licence, and this entry fetched none of them. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimcaches.apps.nvidia.com and nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMCache spec.source.hf.pullSecret, NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret hf-api-secret (NIMCache spec.source.hf.authSecret, NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService reads its model from NIMCache nim-cache-model-free, which this sample file also defines. NIMCache nim-cache-model-free asks the operator to create a 50Gi ReadWriteOnce PersistentVolumeClaim with an empty storageClass, so the storage class is a destination choice. The listing gives the rest. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.

Fixed when this entry was built. These went into the build, so the objects already carry them. To change one, build the entry again from its source.

• source is data/aicr-nim-operator-models/upstream/serving/kserve/standard/multi-llm.yaml
• upstreamCommit is 5d5f66ee0e2e
• sourceFileSha256 is ff06cecd94d3e6e2fc8e7c59718a617b2915efdca60c6bc154a4cc1cd3631b68
• objectSetSha256 is 1de00b1cf1d89ca3a6488a801b445abacea27fe32f814c193a88b46df4b35a93
• image is nvcr.io/nim/nvidia/llm-nim:1.15.0

Yours to supply or check. The destination must have these before the objects are applied. The words in italics say what the Catalog knows about each.

• The k8s-nim-operator chart at >=3.1.0 must be installed first. Every field this sample sets is declared by the apps.nvidia.com/v1alpha1 definitions retained from operator v3.1.0 under examples/aicr/eks-h100-inference-nim/operator-crds. The newest operator release this sample works with is not established. The sample was retained from upstream commit 5d5f66ee0e2e, and this repository does not record which operator release contains that commit. Declared. Not checked on any destination.
• The user supplies this image pull Secret. It is named by NIMCache spec.source.hf.pullSecret and NIMService spec.image.pullSecrets. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMCache spec.source.hf.authSecret. Declared. Not checked on any destination.
• The user supplies this API key Secret. It is named by NIMService spec.authSecret. Declared. Not checked on any destination.
• The NIMService requests 1 GPU per pod as nvidia.com/gpu, with 1 replica. A destination needs that many allocatable GPUs of a kind the image supports, and nothing here checked one. Declared. Not checked on any destination.
• Every object in the sample is in the nim-service namespace. The sample ships no Namespace object, so the namespace must exist before apply. Declared. Not checked on any destination.
• The sample sets spec.inferencePlatform to kserve, so that platform must be installed on the destination before the operator can serve this NIMService. Declared. Not checked on any destination.
• The sample fetches model weights for nvidia/Llama-3.1-Nemotron-Nano-8B-v1 from https://huggingface.co. Those weights carry their own licence, and this entry fetched none of them. Declared. Not checked on any destination.

These are the steps that plain objects cannot carry. ConfigHub stores them and does not carry them out. The words in italics say what the Catalog knows about each.

• Once the destination is known. Install the k8s-nim-operator chart from https://helm.ngc.nvidia.com/nvidia at >=3.1.0 and wait for nimcaches.apps.nvidia.com and nimservices.apps.nvidia.com before applying this sample. The k8s-nim-operator chart is not yet a Catalog entry on this branch, so this requirement names the chart and does not link a record. Who: The platform team that runs the NIM Operator on the selected destination. Generated. Not run on a cluster.
• Once the destination is known. Create the image pull Secret ngc-secret (NIMCache spec.source.hf.pullSecret, NIMService spec.image.pullSecrets) in the sample's namespace before apply. The sample carries the name only, and no credential is held here. Who: The user, with their own NVIDIA entitlement. Generated. Not run on a cluster.
• Once the destination is known. Create the API key Secret hf-api-secret (NIMCache spec.source.hf.authSecret); ngc-api-secret (NIMService spec.authSecret) in the sample's namespace before apply. The sample carries the name only, and no key or token is held here. Who: The user, with their own NVIDIA or model-hub entitlement. Generated. Not run on a cluster.
• Once the destination is known. The NIMService reads its model from NIMCache nim-cache-multi-llm, which this sample file also defines. NIMCache nim-cache-multi-llm asks the operator to create a 50Gi ReadWriteOnce PersistentVolumeClaim with an empty storageClass, so the storage class is a destination choice. The listing gives the rest. Who: The storage owner on the selected destination. Not evaluated.

Test the plain objects yourself. This script tests the retained objects themselves with kubectl, and it needs no cub and no account. Read it first and use a throwaway cluster. The Catalog has no recorded run of this script.

Read what a plan is, and why some entries have one.