Send a test text to the llama.cpp server and verify it responds with a valid embedding vector. If this fails, check that the llama-embed service is running on 127.0.0.1:8081 with --embedding enabled.