Three months ago, I spent 6 hours trying to follow LLM training tutorials. They were either too simple (toy examples) or too complex (assumed I had research labs).

I just wanted to understand how these things work.

So I built create-llm - the tool I wish existed.

🎯 What it does:

Train your own language model in 60 seconds. One command. No setup. Works on any laptop.

🤔 Why it's different:

It doesn't just scaffold code - it TEACHES you. It warns about overfitting, detects config issues, and shows you what went wrong.

The nano template intentionally shows mode collapse with small data. Then you fix it. That's how you learn.

📊 Early traction:

- Live for 1 week

- Active users already training models

- Real-time bug reports and fixes

- Working on M1 Macs, Linux, Windows

🚀 What's next:

Short term: multi-GPU support, better benchmarking

Long term: Cloud training platform (think Vercel for LLMs)

Try it: npx create-llm my-first-llm

I'll be here all day answering questions! Ask me anything about:

- How to train your first model

- Debugging overfitting

- Scaling to larger models

- Contributing to the project

Thanks for checking it out! 🙏

GitHub: https://github.com/theaniketgiri/create-llm

Twitter: https://x.com/theaniketgiri