Also, a question: When you fine-tune a model that had a predefined prompt structure, like, let's say, Mistral for instance with its [INST] Instruction [/INST] do you use that in the fine-tuning process too when you're feeding the new strings?
Fine-tuning Language Models with Predefined Prompt Structures
By
–