fast tokenizer and stream_chat fix

remove unnecessary attention_drop

Update special tokens (#1)

fix import error

support flash attn 2

fix: add eoa into eos_token_id in chat to accelerate chat interface

initial commit internlm2-chat-7b-sft model