Overview A Datawhale course of Jupyter notebooks and write-ups that rebuild LLM components at the PyTorch level without high-level frameworks. Modules cover Tiny Llama3 pretraining, a Transformer, a Qwen2 architecture walkthrough, DDPM diffusion, RAG, GraphRAG, a ReAct-style agent and an evaluation suite. The README is in Chinese, and some modules have companion lecture videos.
Who it is for: Chinese-reading ML learners who want to build LLM, RAG, agent internals
Strengths Covers model, RAG, GraphRAG, agent, eval and diffusion in one repository Code written in plain PyTorch and NumPy rather than wrapped frameworks TinyLLM training is stated to use about 2 GB of VRAM Several modules link to recorded walkthrough videos Weaknesses Documentation is written in Chinese only CC BY-NC-SA 4.0 license forbids commercial use Educational code, not a library or deployable system No releases; each module has its own setup What it needs GPU optional Needs PyTorch, NumPy Models: Qwen2, Llama3 (tiny), DDPM Also in Learning and cookbooks Rank Project Score 1 ai-engineering-from-scratch Hands-on AI engineering curriculum with 542 lessons across 20 phases 91 out of 100 score 2 generative-ai-for-beginners 21-lesson course on building generative AI apps in Python and TypeScript 83 out of 100 score 3 rag_techniques Collection of 42 runnable notebooks implementing RAG techniques 82 out of 100 score # 4cs249r_book Open textbook and course materials for machine learning systems engineering 82 out of 100 score # 5ai-for-beginners 24-lesson AI curriculum with notebooks in PyTorch and TensorFlow (opens in a new tab) 81 out of 100 score
How to choose Check the date: guides older than a year often use APIs that changed. Prefer material with runnable notebooks or repos over slides. Pick by level: curricula start from zero, cookbooks assume you already build. Health
Releases No GitHub releases
Top contributor 100% of commits2 of 2 in the last 12 months
Security advisories NoneOn GitHub, in the last 12 months
OpenSSF Scorecard Not checked Read from GitHub, deps.dev and OpenSSF on 12 Nov 2026. Bots and coding-agent accounts are left out of the commit counts. Not part of the score: how we check .
License in plain words no license No license
Use it at work No
Change it and rebrand it No
Sell it or host it for others No Your part: Without a license you have no right to use, copy or change it. Ask the authors.
A plain summary, not legal advice. Read the license on GitHub ↗ (opens in a new tab)
Where to run it Referral links: if you sign up through one, we get a commission. They never change a rank.
Ask your AI assistant The rankings as an MCP server for Claude, Cursor, ChatGPT and any MCP client. Free, no key, read-only.
Setup for other apps → Claude Code claude mcp add archestack -- npx -y @archestack/mcpCopy
Connector URL https://mcp.archestack.com/mcpCopy
Then ask: Compare tiny-universe with its top alternatives on Archestack
Share this rank #99 of 100 in Learning and cookbooks on Archestack, as of 12 Nov 2026.
Show it in your README or on your site The badge below goes in a README; a card with the rank, score and signals goes on a site or blog. Both update twice a day.
Copy for README Copy for your site
Show code README (Markdown)
[](https://archestack.com/best-of-ai/projects/tiny-universe/)Site or blog (HTML)
<iframe src="https://archestack.com/embed/best-of-ai/tiny-universe/" title="tiny-universe on Archestack" width="400" height="270" style="border:0;max-width:100%" loading="lazy"></iframe>How badges work Preview the card
Best of Open-Source AI on GitHub Each review comes from the project's README and facts we check on GitHub. The fix form opens with tiny-universe filled in.