I want to finetune the bison-32k model for sequential conversation for a fitness session. How do I format my dataset which has the entire fitness session as 1 example. Each conversation has multiple dialogues back and forth. And I have multiple such ...