jingyaogong/minimind
🧠 Train a 64M-parameter LLM from scratch in just 2h!
Build with Backblaze B2
SDKs, agent skills, IDE extensions, and reference pipelines from Backblaze Labs. All open source.
Loading star history...
Where it stands
Rank, growth pace and release cadence relative to everything GStars tracks.
Ranks #7 among AI repositories tracked this week, and #2 over the last 30 days; #469 of 1,950 repositories GStars tracks this week (top 24%).
No releases in the last 90 days; the most recent was 350 days ago, with 19 contributors on GitHub.
Peaked at #2 in the weekly ranking the week of Nov 9, its best in the past year.
Use Cases & Benefits
- Trains ultra-small large language models (LLMs) from scratch in about 2 hours using minimal GPU resources and cost.
- Enables full LLM training pipeline with native PyTorch code, avoiding heavy dependencies and allowing deep understanding and customization.
- Use for researchers or hobbyists wanting to learn LLM internals by training a complete model from zero on a single GPU.
- Use for developers needing a lightweight, cost-effective LLM solution for quick prototyping or domain-specific fine-tuning.
- Use for educators or students seeking an open-source, hands-on tutorial to understand LLM training, fine-tuning, and distillation processes.
Should you depend on it?
License terms, project hygiene and risks, from GitHub metadata and GStars' snapshots.
License
Apache 2.0
Permissive: commercial use, modification and redistribution are allowed with attribution; includes an explicit patent grant from contributors.
Community files
- README: present
- Code of conduct: present
Watch-outs
No watch-outs from GStars' checks.
General guidance from repository metadata, not legal advice. How we score
From the README
"大道至简"
中文 | English
为防止误解，“2小时” 基于NVIDIA 3090硬件设备（单卡）测试，“3块钱” 指GPU服务器租用成本，具体规格详情见下文。
大语言模型（Large Language Model, LLM）的出现引发了全世界对AI的空前关注。 无论是ChatGPT、DeepSeek还是Qwen，都以其惊艳的效果令人叹为观止。 然而，动辄数百亿参数的庞大规模，使得它们对个人设备而言不仅难以训练，甚至连部署都显得遥不可及。 打开大模型的“黑盒子”，探索其内部运作机制，多么令人心潮澎湃！ 遗憾的是，99%的探索只能止步于使用LoRA等技术对现有大模型进行少量微调，学习一些新指令或任务。 这就好比教牛顿如何使用21世纪的智能手机——虽然有趣，却完全偏离了理解物理本质的初衷。 与此同时，第三方的大模型框架和工具库，如transformers+trl，几乎只暴露了高度抽象的接口。 通过短短10行代码，就能完成“加载模型+加载数据集+推理+强化学习”的全流程训练。 这种高效的封装固然便利，但也像一架高速飞船，将我们与底层实现隔离开来，阻碍了深入探究LLM核心代码的机会。 然而，“用乐高拼出一架飞机，远比坐在头等舱里飞行更让人兴奋！”。 更糟糕的是，互联网上充斥着大量付费课程和营销号，以漏洞百出、一知半解的内容推销AI教程。 正因如此，本项目初衷是拉低LLM的学习门槛，让每个人都能从理解每一行代码开始， 从零开始亲手训练一个极小的语言模型。是的，从零开始训练，而不是仅仅进行推理！ 最低只需3块钱不到的服务器成本，就能亲身体验从0到1构建一个语言模型的全过程。 一起感受创造的乐趣吧！
项目包含
希望此开源项目可以帮助LLM初学者快速入门！
Daily snapshots
Stars gained per day, and how forks and open issues moved, from GStars' daily snapshots
+10,210 stars over 90 days across 42 snapshot days; best day +979 on Sep 1.
Forks
8.2K+1.3K
Open issues
64+19
Compared with peers in AI
Fastest-growing repositories GStars tracks in this cohort for the week of Oct 11, by 7-day star growth
| Repository | Stars | +7d | Health | Last release |
|---|---|---|---|---|
| storytold/artcraft | 25.7K | +21.9K | 75 | 17 days ago |
| heygen-com/hyperframes | 62.8K | +4.7K | 45 | today |
| thedotmack/claude-mem | 100.4K | +3.6K | 75 | today |
| calesthio/OpenMontage | 68.0K | +3.4K | 75 | — |
| vxcontrol/pentagi | 28.2K | +2.7K | 75 | 8 days ago |
| jingyaogong/minimind | 63.3K | +0 | 75 | 350 days ago |
| Category median | 62.8K | +3.6K | 75 | — |
Also by jingyaogong
Other repositories from this maintainer that GStars tracks
Discover Repositories
Search across tracked repositories by name or description