从无名小卒到大模型（LLM）大英雄~ 欢迎关注后续！！！
2.2k
Stars
153
Forks
4
Open issues
4
Contributors
AI Analysis
A comprehensive educational resource for learning large language model development from scratch, featuring Jupyter notebooks with hand-written implementations of GPT, MoE models, and fine-tuning techniques (SFT, DPO, RLHF), accompanied by video tutorials on Bilibili. Best suited for ML practitioners and students seeking deep understanding of LLM internals through practical implementation; not for those seeking high-level APIs or production-ready frameworks.
Inferred from signals mentioned in the README (tests, CI, type safety) — not a review of the actual code.
AI's overall editorial judgment — not an average of the bars above, can weigh other factors too.
Chinese-language LLM learning curriculum with hands-on code notebooks and video lectures
LLMs-Zero-to-Hero is a structured educational resource for learning large language model training, covering architecture, pre-training, fine-tuning (SFT, DPO, RLHF), and deployment. Built by a single author in China and distributed via Jupyter notebooks with accompanying Bilibili video lectures. Targets learners who want to understand LLM internals through implementation rather than theory alone. Adoption appears limited to Chinese-speaking communities; limited evidence of enterprise or production usage.
Repository created January 2025 as an educational project. Author explicitly references Andrej Karpathy's pedagogical approach and positions this as a Chinese-language counterpart to similar projects (dive-into-llms, happy-llm). Mirrors broader trend of practitioners creating implementation-focused learning materials in response to proliferation of LLM literature.
Gained 2,234 stars over ~18 months (as of July 2026), with 6 stars in the last 7 days. Growth trajectory suggests niche adoption within specific audience (Chinese-speaking ML enthusiasts, students). Last significant push May 2026 indicates author is still actively developing. Growth rate is modest but steady; not accelerating, not stagnating.
Adoption not verified. No documentation of enterprise usage, benchmark results, or deployment cases. Appears positioned as educational resource rather than production tool. Author links to personal blog, Bilibili channel, and commercial products (ApeRouter), but no metrics on learner volume, course completions, or downstream adoption provided.
Based on README: collection of Jupyter notebooks implementing LLM components from scratch (nanoGPT, MOE models, DeepSeek MLA, GRPO). Likely uses PyTorch for implementation. Appears to follow pedagogical structure (build simple → build complex) rather than production library design.
Not documented in README. No mention of test suites, unit tests, or validation harnesses. Typical for educational notebooks.
Last push 2026-05-04 (59 days before evaluation date) suggests active but not frequent maintenance. README mentions ongoing book-writing project and future video content. No evidence of bug tracking or issue resolution velocity in provided metadata. Single-author project with associated commercial ventures (ApeRouter, ApeCode.ai), which may affect available time.
ADOPT IF: you are a Chinese-speaking learner or researcher seeking hands-on understanding of LLM architectures (dense models, MOE, MLA) and want video-accompanied code notebooks. AVOID IF: you need production-ready libraries, require English-only materials, or seek actively-maintained collaborative frameworks with strong community support. MONITOR IF: the author's book project increases formalization of the curriculum, which could signal transition to larger audience.
Independent dimensions
Mainstream potential
3/10
Technical importance
6/10
Adoption evidence
2/10
- Single-author project with no evident succession planning or collaborative governance — sustainability depends on one individual's availability and interest.
- Primary distribution via Bilibili (Chinese platform) may limit discoverability and contribution from international audience; risk of knowledge fragmentation.
- Educational focus means code is not hardened for production; notebooks may contain inefficiencies or pedagogical shortcuts that do not scale.
- No visible testing, CI/CD, or quality assurance processes documented. Code updates (last push May 2026) do not guarantee correctness.
- Author's attention divided across multiple commercial ventures (ApeRouter, ApeCode.ai); less time for maintenance or responding to issues.
Likely to remain a specialized educational resource within Chinese-speaking LLM enthusiast communities. May grow modestly if book publication succeeds in expanding reach beyond Bilibili audience. Unlikely to become a broadly-adopted reference implementation or to transition to production tool category.
Newsletter
Get analyses like this every Monday
Free weekly digest of the most interesting open-source discoveries.
Languages
Information
- Language
- Jupyter Notebook
- License
- Apache-2.0
- Last updated
- 2mo ago
- Created
- 18mo ago
- Analyzed with
- anthropic/claude-haiku-4-5
Stars over time
Contributors over time
Top 100 contributors only — repos with more will plateau at 100.
Open issues
Open pull requests
No open pull requests.
Top contributors
Recent releases
No releases published yet.
Similar repos
datawhalechina/happy-llm
Happy-LLM is a comprehensive Chinese-language educational project that teaches...
angelos-p/llm-from-scratch
A hands-on educational workshop for learning LLM fundamentals by implementing a...
| Repository | Stars | Week Δ | Language | Score | Updated |
|---|---|---|---|---|---|
|
bbruceyuan/LLMs-Zero-to-Hero
this
|
2.2k | +10 | Jupyter Notebook | 7/10 | 2mo ago |
|
datawhalechina/happy-llm
|
31.9k | — | Jupyter Notebook | 8/10 | 2mo ago |
 Lordog/dive-into-llms
|
42.3k | — | Jupyter Notebook | 7/10 | 9mo ago |
 nndl/llm-beginner
|
6.5k | — | Python | 8/10 | 2w ago |
|
angelos-p/llm-from-scratch
|
3.1k | — | — | 8/10 | 2mo ago |
|
datawhalechina/llms-from-scratch-cn
|
4.3k | — | Jupyter Notebook | 7/10 | 4mo ago |
Significantly larger audience (18.4x more stars). LLMs-Zero-to-Hero appears to offer more structured curriculum and video accompaniment; dive-into-llms may have broader implementation coverage.
More established community project with institutional backing. LLMs-Zero-to-Hero is single-author with stronger emphasis on implementation-from-scratch pedagogogy.
Similar scope and star count. LLMs-Zero-to-Hero emphasizes hands-on coding; llm-beginner may emphasize theory. Both appear to serve similar niche.
Comparable adoption to LLMs-Zero-to-Hero. Likely similar teaching materials format. Direct competitors for mindshare in Chinese learner communities.
Smaller but overlapping niche. Collaborative community project vs. LLMs-Zero-to-Hero's single-author approach may affect sustainability.
