minimind
by jingyaogong
MiniMind is a set of Python code and open datasets for training a very small language model (about 64 million parameters) from scratch. It is mainly a hands-on way to learn how language models are built, trained and fine-tuned, and it also includes a simple chat interface and an OpenAI-compatible server.
- Stars
- 64.5k
- Licence
- Apache-2.0
- Runs on
- A gpu machine, with a GPU
- Setup
- Technical
- Language
- Python
- Age
- 822 days
What you can do with it
- Train a small language model from scratch, including pretraining and fine-tuning, on a single consumer GPU
- Fine-tune the model on your own data using LoRA or distillation
- Run the trained model through a chat web page or an OpenAI-compatible API server
Make money with it
Someone with machine learning skills could offer custom small-model fine-tuning or training workshops, using this as the teaching base.
An idea, not financial advice. Check the licence before charging for anything built on it.
Ways a business could use it
- Build a small, private model trained on your own domain data for simple tasks without paying per-use API fees
- Use it to train staff or a technical team in how language models work before deciding on larger AI projects
Can you use it commercially?
Free to use, change and sell, including in paid products. Keep the copyright notice.
Not legal advice. Read the project's licence file before building a product on it.
Getting it running
You clone the repository, install the Python dependencies with pip, download the datasets and run the training scripts, ideally on a GPU such as an NVIDIA 3090.
Follow the official instructions in the project README.
Things to check
- The model is tiny, so its answers are far weaker than commercial chat models
- Documentation is mainly in Chinese, with an English version linked separately
- Training needs a capable GPU, either owned or rented
Who's talking about it
Checked 27 Oct 2026. Summary drafted by AI from the project's README.
