▲ 687 today66.8k starsPythonHN ×1
Train a 64M LLM from scratch in 2 hours for $3
What is minimind?
MiniMind lets you build a 64-million-parameter language model from zero using only a few dollars of cloud GPU time. The whole pipeline, from tokenizer to RL-fine-tuning, is provided in pure PyTorch.
Who it is for
If you want to understand how large language models work under the hood, this gives you a cheap, hands-on playground.
Try it in 60 seconds
You need: A machine with an NVIDIA GPU (e.g., 3090) and Python installed
No one-line install. The README has the setup steps.
Apache-2.0 licenseUpdated yesterdayPython63 open issues
New projects are not vetted. Skim the code, and run it in a fresh folder or a container before giving it your files or keys.
The story so far
Early Signal first spotted jingyaogong/minimind on November 19, 2026, when it had 66.6k GitHub stars. It now has 66.8k. On its busiest day it gained 687 stars. It is published under the Apache-2.0 license and has 8.8k forks. It has been discussed in 1 Hacker News thread, the top one with 3 points. Early Signal files it under LLMs, Inference, Agents.
Questions about minimind
- Is minimind free and open source?
- Yes. jingyaogong/minimind is open source under the Apache-2.0 license, and the code is on GitHub at https://github.com/jingyaogong/minimind.
- What do I need to run minimind?
- A machine with an NVIDIA GPU (e.g., 3090) and Python installed