TroyZ

Fine-tuning is probably not the way to do it.

Try embedding, semantic search, retrieval, and plugging the relevant parts into the prompt.

You may need: - summarizer prompt to summarize your project structure, main functions, methods. - vector store/database to store and retrieve your relevant code from code base - coder prompt to write code based on the retrieved part.

Check out langchain: https://langchain.readthedocs.io/

For a high level overview of transformers/GPT, you can check out this article: https://medium.com/design-bootcamp/how-chatgpt-really-works-...

The article provides references to the original papers and other articles explaining the subject, which can be great sources to dive deeper.

You can actually easily incorporate activities in your daily routines. Get a standing desk, do some quick stretches/body-weight exercise after 40 min of coding, march/bike in place while watching videos/TV, stroll afer meals, etc.

Getting a pedometer app also helps. If you're self-monitoring your activity level, there's a great chance you'll try to improve it (consciously or unconsciously).[1]

Allow me to advertise our app: Pacer Pedometer[2]. It comes with activity tracking, fun virtual activity challenges, healthy lifestyle lessons, guided body-weight exercise etc. We also have a corporate plan in which you can join challenges with your colleagues.

[1] Kang, Minsoo, et al. "Effect of pedometer-based physical activity interventions: a meta-analysis." Research quarterly for exercise and sport 80.3 (2009): 648-655. [2] https://apps.apple.com/us/app/pacer-pedometer-step-tracker/i...