Agentic Translation

Agentic Translation#

Overview#

agentic-translation UI

This Solution Blueprint illustrates how language translation can be implemented using AIMs. It uses agentic translation, employing multiple LLM agents working collaboratively, where models critique, evaluate, and refine each other’s outputs to improve the overall quality of the translation.

The blueprint follows a trilateral collaboration framework from Wu et al. with an Action agent, Critique agent, and Judgment agent iteratively contributing to the translation task until the Judgment agent approves the output. The prompting strategy is adapted from Andrew Ng’s example.

AMD Solution Blueprints are packaged as Helm charts for deployment on a Kubernetes cluster. For development or further exploration, the source code is public and available in the Solution Blueprints GitHub repository.

Architecture#

The machine translation is performed by a collaboration of three LLM agents. The machine translation is performed by a collaboration of three LLM agents.

The blueprint provides a Streamlit web application for agentic translation with multi-agent LLM collaboration. By default, an AIM is deployed (Llama-3.3-70B) to power the Action, Critique, and Judgment agents.

Component

Role

Streamlit UI

Web interface for entering text, instructions, and languages; reviewing translations and agent dialogue

Agentic translation pipeline

Action, Critique, and Judgment agents with iterative refinement

AIM LLM

Powers the multi-agent translation workflow (default: Llama-3.3-70B)

LangChain

Implementation of the agents

Key Features#

  • Additional instructions: The agentic translation can incorporate additional user-provided instructions

  • Multilingual support: Users can freely use any languages supported by the underlying LLM

  • Visible agent dialogue: The conversation between the agents is made visible to the user

  • Long inputs: If the input text exceeds the model’s context window, it is automatically split into smaller chunks for processing

Getting Started#

This is a quick start guide on how to deploy the blueprint. For advanced options, such as reusing an existing AIM, providing a Hugging Face token, or overriding storage classes, see Deploying Solution Blueprints with Helm or explore the advanced deployment guide.

Prerequisites#

System Requirements#

This blueprint can be deployed on AMD Instinct (default) and AMD Radeon. The blueprint requires the following cluster resources by default, depending on the hardware being used:

Resource

Instinct

Radeon

GPUs

1

1

CPUs

5 CPU cores

5 CPU cores

RAM

68 Gi

36 Gi

To deploy to the Kubernetes cluster, ensure the following prerequisites are met:

  • kubectl: Installed and configured to communicate with the cluster

  • Helm 3.17 or higher: Installed on your local machine

Deployment#

For advanced deployment options, explore the advanced deployment guide. Solution Blueprints are packaged as OCI-compliant Helm charts in the Docker Hub registry and can be deployed to a Kubernetes cluster with a single command. Define the name (deployment name) and the namespace (Kubernetes namespace), then pipe the output of helm template to kubectl apply -f -.

Find the deployment command below.

Note: You can create a namespace using kubectl create namespace <my-namespace>.

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

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

Verify Deployment#

To check the status of the deployment, run:

kubectl get pods -n $namespace

Wait until all pods report Running and Ready.

Connect to UI#

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

kubectl port-forward services/aimsb-agentic-translation-${name} 8501:8501 -n $namespace

Once connected, use the application as follows:

  1. Enter the source text and optional extra instructions

  2. Choose source and target languages (any pair supported by the underlying model)

  3. Run translation and review the final output and the visible agent conversation

Clean Up#

When you are finished, remove the deployed resources using the same deployment command, with kubectl delete instead of kubectl apply. For example, for Instinct use the following command:

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

Third-Party Components#

This Solution Blueprint uses multiple third-party components. To see the full set of software and Python dependencies, explore the repository source and dependency files. The table below highlights some of the key components. For further license information, refer to each component’s official documentation.

Component

License

Streamlit

Apache 2.0

LangChain

MIT

translation-agent (andrewyng/translation-agent)

MIT

  • Translation Agent: Agentic translation using reflection workflow

    • Original source: https://github.com/andrewyng/translation-agent

    • License: MIT https://github.com/andrewyng/translation-agent?tab=MIT-1-ov-file

Terms of Use#

AMD Solution Blueprints are released under the MIT License, which governs the parts of the software and materials created by AMD. Third-party Software and Materials used within the Solution Blueprints are governed by their respective licenses.

Contents