Agentic Translation Deployment Guide

Agentic Translation Deployment Guide#

Solution Blueprints are provided as Helm Charts. The recommended approach to deploy them is to pipe the output of helm template to kubectl apply -f -. We don’t recommend helm install, which by default uses a Secret to keep track of the related resources. This does not work well with Enterprise clusters that often have limitations on the kinds of resources that regular users are allowed to create.

This blueprint supports AMD Instinct (default) and AMD Radeon platforms. Unless otherwise specified, the commands below cover the default Instinct deployment. For deployment with Radeon, see:

Multi-platform Support#

The chart ships defaults for two platforms, selected with --set global.platform=<platform>: instinct (GPU, the default) and radeon (GPU). Each sets a matching AIM image and resource profile; inspect them with helm show values . --jsonpath '{.llm.platformDefaults}'.

Helm note: Built and tested on Helm 3.17 or higher. On Helm v4, if the piped kubectl apply is rejected, run helm pull oci://registry-1.docker.io/amdenterpriseai/aimsb-agentic-translation --untar first and template the local ./aimsb-agentic-translation directory instead.

AMD Instinct (GPU, default)#

To deploy the blueprint, run the following command:

name="my-deployment"
namespace="my-namespace"
helm template $name oci://registry-1.docker.io/amdenterpriseai/aimsb-agentic-translation \
  | kubectl apply -f - -n $namespace

AMD Radeon (GPU)#

To deploy the blueprint, run the following command:

name="my-deployment"
namespace="my-namespace"
helm template $name oci://registry-1.docker.io/amdenterpriseai/aimsb-agentic-translation \
  --set global.platform=radeon \
  | kubectl apply -f - -n $namespace

Using an existing deployment or external LLM#

By default, any required AIMs are deployed by the helm chart. If you already have a compatible AIM deployed, you can use that instead and reuse resources.

To use an existing deployment or external LLM, set the value llm.existingService to that endpoint. Then, any other values you pass in the llm mapping are simply ignored, and your existing service is used instead. You should use the Kubernetes Service name, or if the service is in a different namespace, you can use the long form <SERVICENAME>.<NAMESPACE>.svc.cluster.local:<SERVICEPORT>. If needed, you can pass a whole URL.

Full example command:

name="my-deployment"
namespace="my-namespace"
servicename="aim-llm-my-model-123456"
helm template $name oci://registry-1.docker.io/amdenterpriseai/aimsb-agentic-translation \
  --set llm.existingService=$servicename \
  | kubectl apply -f - -n $namespace

API Key and Model Configuration for External LLM#

When using an external LLM service, you can optionally configure the API authentication credentials and specify a particular model:

  • llm.apiKey (optional): Bearer token for API authentication

  • llm.model (optional): The specific model identifier to use (e.g., openai/gpt-oss-20b, gpt-4-turbo)

Specifying the model is not needed when using AIMs, as the instance typically only serves one model. If llm.model is left empty, the list of models is queried from the API, and the first available model is used.

Example command:

name="my-deployment"
namespace="my-namespace"
api_url="https://llm-api.example.com"
api_key="<YOUR_API_KEY>"
model_name="openai/gpt-oss-20b"

helm template $name oci://registry-1.docker.io/amdenterpriseai/aimsb-agentic-translation \
  --set llm.existingService=$api_url \
  --set llm.apiKey=$api_key \
  --set llm.model=$model_name \
  | kubectl apply -f - -n $namespace

Using Kubernetes Secrets for API Key#

For enhanced security, you can store the API key in a Kubernetes secret and reference it:

name="my-deployment"
namespace="my-namespace"
secretname="my-secretname"
api_url="https://llm-api.example.com"

# Create the secret
kubectl create secret generic $secretname -n $namespace \
  --from-literal=api-key="<YOUR_API_KEY>"

# Deploy with secret reference
helm template $name oci://registry-1.docker.io/amdenterpriseai/aimsb-agentic-translation \
  --set llm.existingService=$api_url \
  --set llm.apiKeySecretRef.name=$secretname \
  --set llm.apiKeySecretRef.key=api-key \
  | kubectl apply -f - -n $namespace

Default AIM image and GPU compatibility#

By default, the chart deploys Meta Llama 3.3 70B with this AIM: amdenterpriseai/aim-meta-llama-llama-3-3-70b-instruct:0.11.1

On newer GPUs, this default image may not be the best match and can fail to start or run sub-optimally. To choose a newer AIM or deploy a different LLM, override llm.image to a compatible image. See the catalog of available AIMs for options.

Example:

name="my-deployment"
namespace="my-namespace"
helm template $name oci://registry-1.docker.io/amdenterpriseai/aimsb-agentic-translation \
  --set llm.image=amdenterpriseai/aim-meta-llama-llama-3-3-70b-instruct:<NEWER_TAG> \
  | kubectl apply -f - -n $namespace

Using a custom image and imagePullSecrets#

To use a custom version of the image from another repo, you can set the image and imagePullSecrets, if necessary, from the command line as follows:

name="my-deployment"
namespace="my-namespace"
imagerepository="my-repo/custom-agentic-translation-application"
imagetag="0.0.1"
secretname="my-registry-secret"
helm template $name oci://registry-1.docker.io/amdenterpriseai/aimsb-agentic-translation \
  --set image.repository=$imagerepository \
  --set image.tag=$imagetag \
  --set imagePullSecrets[0].name=$secretname \
  | kubectl apply -f - -n $namespace

Connecting#

Option 1: Port Forwarding#

To connect to the UI, port-forward port 8501. The UI will then be available at http://localhost:8501.

kubectl port-forward service/aimsb-agentic-translation-$name 8501:8501 -n $namespace

Option 2: HTTPRoute (Gateway Access)#

If your cluster has a Gateway API compatible gateway (e.g., Kubernetes Gateway, Istio, etc.), you can enable HTTPRoute creation to route traffic through the gateway.

Prerequisites:

  • A Gateway named https must exist in the envoy-gateway-system namespace (or configure a different gateway)

  • The Gateway must be properly configured with listeners

Enabling HTTPRoute:

Use --set http_route.enabled=true in the helm template command to enable HTTPRoute creation:

name="my-deployment"
namespace="my-namespace"
helm template $name oci://registry-1.docker.io/amdenterpriseai/aimsb-agentic-translation \
  --set http_route.enabled=true \
  | kubectl apply -f - -n $namespace

Obtaining the URL:

The URL to access the blueprint via HTTPRoute is formed by the service name and the hostname of the gateway. Use this command to produce the URL by querying the hostname from the cluster:

echo "https://aimsb-agentic-translation-$name$(kubectl get gtw -A -o jsonpath='{.items[*].spec.listeners[?(@.name=="https")].hostname}' | tr -d \*)/"

Clean Up#

When you are finished, remove the deployed resources:

helm template $name oci://registry-1.docker.io/amdenterpriseai/aimsb-agentic-translation \
  | kubectl delete -f - -n $namespace
Contents