embedding-model-clients

This project provides a collection of clients for interacting with various embedding models. It includes modules for specific embedding model providers like Graphwise Transformer and OpenAI API, including Azure OpenAI deployments. Clients are implementations of langchain4j EmbeddingModel interface. They can be used in systems that provide similarity search by creating embeddings from texts, such as GraphDB Elasticsearch and Opensearch Connectors. Existing clients serve as examples and default implementations that GraphDB connectors use to provide similarity searches. You can provide additional clients by implementing EmbeddingModel interface and add them to your GraphDB distribution to use for similarity searches in GraphDB Connectors.

Modules

• assembly: Create an assembly jar that contains clients and their dependencies.
• embedding-clients-common: Common code used by the other client modules.
• graphwise-transformer-client: A client for interacting with a Graphwise Transformer.
• openai-embedding-client: A client for interacting with the OpenAI embedding API.
• azure-embedding-client: A client for interacting with the Azure OpenAI embedding API.
• aws-bedrock-embedding-client: A client for interacting with the Aws Bedrock embedding API.

Building

To build the project, you can use Maven. Run the following command from the project's root directory:

mvn clean install

Installation

1. Build the project to create the assembly JAR.
2. Copy the generated JAR from ./assembly/target/embedding-model-clients-assembly-{project.version}.jar to your application's classpath.

• For GraphDB Connectors, place the JAR in the directory of the respective connector, for example: dist/graphdb/target/graphdb/lib/plugins/elasticsearch-connector/.

Configuration

Clients are configured via system properties.

GraphwiseTransformerClient

Property	Description	Required	Default
graphwise.transformer.address	The host and port of the GraphWise Transformer service.	no	localhost:5050
graphwise.transformer.embedding.model.name	The name of the sentence transformer model to use.	no	sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2
graphwise.transformer.batch.size	The maximum request batch size in kilobytes.	no	256
graphwise.transformer.auth.token.secret	Shared secret for authentication.	no	none
graphwise.transformer.thread.pool.size	The size of the client-side thread pool.	no	Number of available processors

OpenAIEmbeddingClient

Property	Description	Required	Default
openai.embedding.model.api.key	Your OpenAI API key.	yes	none
openai.embedding.model.name	The OpenAI model to use.	no	none
openai.embedding.model.dimensions	The OpenAI model dimensions.	no	none
openai.embedding.model.baseUrl	The base URL for connecting.	no	none

AzureEmbeddingModel

Property	Description	Required	Default
azure.embedding.model.api.key	Your Azure API key.	no	none
azure.embedding.client.id	Specifies the client ID of Microsoft Entra app to be used.	no	none
azure.embedding.tenant.id	Specifies the tenant ID.	no	none
azure.embedding.client.secret	Specifies the client secret.	no	none
azure.embedding.model.name	The Azure deployment to use.	no	none
azure.embedding.model.dimensions	The Azure deployment dimensions.	no	none
azure.embedding.model.baseUrl	The base URL for connecting.	yes	none

Environment variables	Description	Required	Default
AZURE_CLIENT_ID	Specifies the client ID of Microsoft Entra app to be used.	no	none
AZURE_TENANT_ID	Specifies the tenant ID.	no	none
AZURE_CLIENT_SECRET	Specifies the client secret.	no	none

Authentication workflow:

• If azure.embedding.model.api.key is provided, then this key is used for authentication.
• If the key is not provided, then the workflow tries to use the properties azure.embedding.client.id, azure.embedding.tenant.id, and azure.embedding.client.secret.
• If neither the key nor the properties are provided, the workflow falls back to DefaultAzureCredentialBuilder (i.e., using the environment variables AZURE_CLIENT_ID, AZURE_TENANT_ID, and AZURE_CLIENT_SECRET).

Note: Keep in mind that GraphDB backups also use these environment variables as the preferred option.

AwsBedrockEmbeddingModel

Property	Description	Required	Default
aws.bedrock.embedding.model.name	The AWS model.	no	none
aws.bedrock.embedding.model.region	Your AWS region.	yes	none

Additionaly, for AWS access 2 environment variables must be defined:

Environment variable	Description	Required
AWS_ACCESS_KEY_ID	Your AWS Key ID.	yes
AWS_SECRET_ACCESS_KEY	Your AWS Secret Key.	yes

Usage

Once the JAR is on the classpath and the necessary properties are configured, you can use the clients in GraphDB Connectors by specifying the fully qualified class name of the desired implementation in the connector configuration.

Example values for embeddingModel parameter:

• com.ontotext.embeddings.GraphwiseTransformerClient
• com.ontotext.embeddings.OpenAIEmbeddingModel
• com.ontotext.embeddings.AzureEmbeddingModel
• com.ontotext.embeddings.AwsBedrockEmbeddingModel

License

Licensed under Apache 2.0.