Test mocking fails for Xinference Client

[slobentanzer]

@fengsh27 on the current PR #102, CI fails because the Client is not found in the llm_connect.py module. If I remember correctly, you moved the import into the XinferenceConversation to keep the server lightweight, right?

https://github.com/biocypher/biochatter/blob/2982143010e8e268207fa6dc8af8bb50ecfdd108/biochatter/llm_connect.py#L412

Is this still necessary, so do we need to think about how to adjust the test so it does not fail, or should we rather move the Client import back to the top?

AttributeError: <module 'biochatter.llm_connect' from '/home/runner/work/biochatter/biochatter/biochatter/llm_connect.py'> does not have the attribute 'Client'

[fengsh27]

@slobentanzer As we had moved import xinference to Constructor, I imported xinference.client.Client at the top of test_llm_connect.py. So, we need to mock "xinference.client.Client" instead:

@pytest.fixture
def xinference_conversation():
    with patch("xinference.client.Client") as mock_client:   # mock xinference
        mock_client.return_value.list_models.return_value = xinference_models
        mock_client.return_value.get_model.return_value.chat.return_value = (
            {"choices": [{"message": {"content": "Human message"}}]},
            {"completion_tokens": 0},
        )
        conversation = XinferenceConversation(
            base_url="http://llm.biocypher.org",
            prompts={},
            correct=False,
        )
    return conversation

See https://github.com/biocypher/biochatter/blob/2982143010e8e268207fa6dc8af8bb50ecfdd108/test/test_llm_connect.py#L172

[slobentanzer]

makes sense, thanks :)