* [LLM] Support CPU Deepspeed distributed inference * Update run_deepspeed.py * Rename * fix style * add new codes * refine * remove annotated codes * refine * Update README.md * refine doc and example code