Add GRPO training example and support vllm in inference script (#12)

* Add GRPO Training support

* Add vllm inference support

* Update format and readme.

* fix format issue

* update path

* update default eos_token as true