Creating Content for Blogs

API
,
1

I am currently learning about fine-tuning models, and my goal is to generate blog articles using a fine-tuned model. To achieve sufficient personalization and the ability to mimic my writing style, I hope the model can fully learn from my sample data during the fine-tuning process. Therefore, I want to assist the model by providing detailed annotations on the articles, such as the structure of the article, the purpose of different paragraphs, the writing methods used, new vocabulary, etc.

My questions are:

1.Do I need to do this?
2.Does the dataset format support custom fields? If not, can I embed this information into the prompt and completion fields? Please see my example.
3.Which base model should I choose for fine-tuning with the goal of writing?
4.How should I represent numbered lists or bullet points within the articles?

Thank you all for taking the time to review my immature and somewhat naive questions. Since you have watched, please share your suggestions.

{
  "prompt": "Title: The Application and Future of AI in Education\nStructure: Introduction, Body, Conclusion\nSection: Introduction\nPurpose: Introduce the background and purpose of the article\nWriting Methods: Background introduction, Problem posing\nParagraph:",
  "completion": "With the rapid development of artificial intelligence technology, the application of AI in various fields is becoming more and more extensive, and education is no exception. This article will explore the current application and future prospects of AI in education."
}
{
  "prompt": "Title: The Application and Future of AI in Education\nStructure: Introduction, Body, Conclusion\nSection: Body\nPart: Current Applications\nPurpose: Discuss the specific applications of AI in education\nWriting Methods: Data analysis\nParagraph:",
  "completion": "AI currently has many practical applications in education, such as personalized learning, intelligent tutoring systems, and automated assessment."
}
{
  "prompt": "Title: The Application and Future of AI in Education\nStructure: Introduction, Body, Conclusion\nSection: Body\nPart: Future Developments\nPurpose: Explore the future prospects of AI in education\nWriting Methods: Future outlook\nParagraph:",
  "completion": "In the future, AI will play a greater role in education, including the widespread use of virtual teachers and smart classrooms."
}
{
  "prompt": "Title: The Application and Future of AI in Education\nStructure: Introduction, Body, Conclusion\nSection: Conclusion\nPurpose: Summarize the main points of the article\nWriting Methods: Summary\nParagraph:",
  "completion": "Overall, the application prospects of AI in education are broad and will greatly improve the quality and efficiency of education."
}

@N2U @curt.kennedy @PaulBellow @Foxalabs

If you have tackled similar issues, I hope you can provide the dataset format, scripts, etc.

1 Like
2

Of course you can, but keep in mind that we haven’t fine-tuned GPT-4 yet. Also, you can do this properly in JSON format for the datasets. Actually, you can put the points that will differentiate your writing into the fine-tuned JSON file as the base model. For example, if you talk more about examples or descriptions while writing on a topic, you can put this into the fine-tuned JSON file. Good luck!

1 Like
3

OK, thanks. Which base model is usually better for article generation?

4

You could play around with GPT-3.5-turbo-16k, it is low on cost too. Set params like temp or top_p, either of those during completions.

5

Ok , thank you very much！