Early tests by Google Cloud using llm-d show 2x improvements in time-to-first-token for use cases like code completion, ...