1) Move LM input/response tagging logic from LangFunc to LanguageModel.
2) Improve debugging, preferred prompt.formatted_text over prompt.text when available.
3) Refine the logic to attach score/logprobs/usage to response message, which is now more general.
Refine
lf.LanguageModel.1) Move LM input/response tagging logic from LangFunc to LanguageModel. 2) Improve debugging, preferred
prompt.formatted_textoverprompt.textwhen available. 3) Refine the logic to attach score/logprobs/usage to response message, which is now more general.