Simple enough. Framing this slightly differently - rather than tasking the model to merely follow instructions, I sometimes put it in a "question-answer" mode instead, since much of GPT4All's dataset is about answering questions. Admittedly, not every entry in the dataset is a question, but I'm sure some overlap between "follow instructions" and "generate answers to questions" is fine. I can remove said prompts if needed.
Simple enough. Framing this slightly differently - rather than tasking the model to merely follow instructions, I sometimes put it in a "question-answer" mode instead, since much of GPT4All's dataset is about answering questions. Admittedly, not every entry in the dataset is a question, but I'm sure some overlap between "follow instructions" and "generate answers to questions" is fine. I can remove said prompts if needed.