How Far Are We to GPT-4V? Closing the Gap to Commercial Multimodal Models with Open-Source Suites