Test Execution

MacOS or Linux without nvidia graphic card

If ollama is already installed on the system execute the test with

bash

mvn verify -Dollama.endpoint=http://localhost:11434/ -Dollama.model=granite4:3b -Dollama.instance.type=remote

The Ollama docker image is really slow on macbook without nvidia hardware acceleration

Linux with Nvidia graphic card

The hardware acceleration can be used, and the test can be executed with

bash

mvn verify -Dollama.container.enable.gpu=enabled

Embedding tests

The embedding integration tests require an embedding model. By default, the test infrastructure uses granite-embedding:30m.

To run embedding tests with a local Ollama instance:

bash

mvn verify -Dollama.endpoint=http://localhost:11434/ -Dollama.model=granite4:3b -Dollama.embedding.model=granite-embedding:30m -Dollama.instance.type=remote

Make sure the embedding model is pulled in Ollama before running the tests:

bash

ollama pull granite-embedding:30m

Running with OpenAI

To run tests against OpenAI API instead of Ollama:

bash

mvn verify -Dollama.instance.type=openai -Dopenai.api.key=sk-xxx -Dopenai.model=gpt-4o-mini -Dopenai.embedding.model=text-embedding-ada-002

Or using environment variables:

bash

export OPENAI_API_KEY=sk-xxx
export OPENAI_MODEL=gpt-4o-mini
export OPENAI_EMBEDDING_MODEL=text-embedding-ada-002
mvn verify -Dollama.instance.type=openai