# datawhalechina/tiny-universe

《大模型白盒子构建指南》：一个全手搓的Tiny-Universe

Repository: https://github.com/datawhalechina/tiny-universe
Canonical: https://ross.abutalabs.com/products/tiny-universe
Language: Jupyter Notebook
License Family: other
Topics: rag, agent, diffusion, evaluation-metrics, llama, qwen, transformers
Last push: 2026-02-12T02:00:08+00:00

## Health v2 (maintenance only)
Score: 55/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 67, release rhythm 35, longevity 62
- inputs: {"age_days": 879, "days_push": 203, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 5028, forks 476 (observed 2026-08-28T04:09:04.934464+00:00)

## What it is
An open-source Chinese-language tutorial series ('Tiny-Universe') that teaches building large language model systems from scratch, covering LLM internals, RAG, Agent, evaluation, diffusion, and GraphRAG with full PyTorch-level implementations. It is structured as Jupyter Notebook content for learners who want to understand LLM principles beyond high-level frameworks.

## Use cases
- learn how transformers and LLMs work internally by implementing them from scratch
- build a minimal RAG framework by hand to understand retrieval-augmented generation
- implement a minimal agent system without high-level frameworks
- pretrain a tiny Llama3-style model on limited GPU memory
- understand how to evaluate large language models
- learn diffusion models by coding an image generation model from zero
- study GraphRAG by building a simplified version

## When to choose
- you already know deep learning basics and want to understand LLM internals at the PyTorch level
- you prefer hands-on coding over using packaged APIs and frameworks
- you want a full-stack LLM curriculum covering model, RAG, agent, and evaluation
- you learn best by reproducing systems from principles with annotated code

## When to avoid
- you need a production-ready RAG, agent, or evaluation framework
- you want a quick-start guide using high-level APIs like LangChain or OpenAI
- you need English-language documentation
- you need a maintained software library with a license for commercial use

## Facets
- artifact type: learning-resource
- maturity: active
- function: machine-learning, deep-learning, llm-training, rag, agent-framework, data-science
- domain: large-language-models, tutorials, artificial-intelligence, deep-learning
- platform: python
- tags: educational, from-scratch, hands-on, jupyter-notebooks, transformer, diffusion-models, evaluation, graphrag, chinese, retrieval-augmented-generation, ai-agents

## Member repositories
- datawhalechina/tiny-universe (main) score 55

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:09:04.934464+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T18:17:28.161689+00:00, confidence not recorded.
  - readme: https://github.com/datawhalechina/tiny-universe (fetched 2026-08-28T04:09:04.934464+00:00, sha 0d7881178bd2)
- Data as of 2026-08-30T08:39:29.467469+00:00.