Add gpt-2 tokenizer, pytorch artifacts, trained model

initial commit