Daily-LLM is a Chinese-first, bilingual knowledge base for understanding how deep-learning and large-model architectures evolved. Markdown remains the canonical content source; the Web app turns that material into a searchable timeline, family browser, and interactive learning experience.

Timeline: 19 Architecture Families

#	Family	Years	Focus
01	Convolutional Neural Networks	1998–2022	Learned visual features and deep residual backbones
02	RNN / LSTM / GRU	1986–2015	Recurrent memory and sequence modeling
03	Word Embeddings	2013–2018	Distributed and contextual word representations
04	Generative Adversarial Networks	2014–2018	Generative modeling through adversarial training
05	Transformer	2017–2022	Parallel sequence modeling with attention
06	BERT Family	2018–2019	Bidirectional pretraining and task fine-tuning
07	GPT and Scaling	2018–2023	Autoregressive pretraining and scaling laws
08	Vision Transformers	2020–2022	Transformer architectures for vision
09	Multimodal Alignment	2021–2023	Shared representations across images and language
10	Diffusion Models	2020–2023	High-quality generation through iterative denoising
11	Parameter-Efficient Fine-Tuning	2019–2023	Adapting large models with a small trainable footprint
12	Alignment and RLHF	2020–2023	Aligning model behavior with human preferences
13	Mixture of Experts	2017–2024	Separating parameter capacity from active compute
14	RAG and Agents	2020–2023	Connecting models to knowledge and tools
15	Reasoning Models	2022–2025	Scaling inference-time computation
16	World Models and Video Generation	2018–2024	Learning environment dynamics and generative worlds
17	Graph Neural Networks	2017–2021	Learning representations over arbitrary graphs
18	Speech and Audio Models	2020–2023	Self-supervised speech and audio-token generation
19	Recommendation Systems	2016–2018	Feature interactions, user interest, and graph recommendation

Browse all 89 works by year in TIMELINE.md, or review shared prerequisites in foundations/.

Interactive Web App

The app in web/ provides:

• a density-aware chronological timeline;
• a 19-family visual browser;
• search across model names and key ideas;
• Markdown article rendering with math, code, diagrams, and local assets;
• interactive “golden sample” pages for selected models;
• shared foundation modules.

The public deployment URL will be added after the first release. To run it locally:

cd web
npm install
npm run dev -- --host 127.0.0.1 --port 5173 --strictPort

Open http://127.0.0.1:5173/.

Repository Structure

• 01-cnn/ … 19-recommendation/ — canonical family and node content
• foundations/ — shared foundations such as optimization, normalization, and attention
• projects/ — cross-family capstone projects
• web/ — React and TypeScript Web experience
• TIMELINE.md — generated chronological index; do not edit manually
• _archive/ — read-only historical material

Quick Start

git clone https://github.com/zkywsg/Daily-LLM.git
cd Daily-LLM
pip install -r requirements.txt

When node frontmatter changes or a node is added, regenerate the timeline and Web metadata:

python3 scripts/generate_timeline.py

Contributing

Contributions that improve explanations, examples, diagrams, facts, accessibility, or the Web experience are welcome. Read CONTRIBUTING.md before submitting changes.

License

Daily-LLM is released under the MIT License.