LLM Quantization with Global Mixed-precision between Output-features and Highly-efficient System Design - View it on GitHub