CoT in PPO format?
#15
by
laelhalawani
- opened
Hello again,
What's the format of the CoT prompt output with the PPO model.
Does it use custom xml styled outputs or any special tokens to denote translated part from the CoT part?
@laelhalawani
The output of CoT prompt is This is the translation text. \n\n[COT] This is the explanation. Thus, you can use [COT] to extract the translation. If the model considers the input sentence to be relatively simple and does not require CoT, it will directly output the translated result without [COT].
Super, thank for your answer!