llm-beginner

What it is

LLM-Beginner is a Python-based, MIT-licensed, open repository described as an independent beginner tutorial for large models and agents. It comprises six tasks that progress from implementing a transformer component to building an agent that can call tools, and includes a companion book released as a PDF.

How it works

The project provides task directories (task-1 to task-6) each containing a requirements.txt, data/download.py, eval/run.py, and tutor_prompt.md for code review prompts. Implementations are expected under each task's src/ with exports following a specified interface. Self-check scripts evaluate key contracts and write results to eval/result.json. It uses a local, task-scoped progression, with dependencies and data downloads handled per task.

Getting started

• Environment setup per task: install dependencies listed in task-1-transformer/requirements.txt (example shown in README).
• Data download and self-check: follow the standard flow: download data, implement src/, run eval/run.py to produce results.
• Command examples provided in README include:

  • cd task-1-transformer
    python data/download.py        # 1. 下载数据 / 模型（命令见下表，部分任务带参数）
    # 2. 在 src/ 下写好你的实现（见本任务 README「实现约定」）
    python eval/run.py             # 3. 跑自检，结果写入 eval/result.json
    

-HF endpoint proxy and optional data sources are described, including environment variable usage for HF_ENDPOINT.

Recent releases

• Latest: book-llm-agent 大模型与智能体 —— PDF 全书 (2026-05-28)
  • This release updates the llm-agent.pdf via release-pdf.ps1 with a fixed URL: https://github.com/nndl/llm-beginner/releases/download/book-llm-agent/llm-agent.pdf

Traction

• Stars: 6604

Behind the repo

• Publisher/creator chain not explicitly stated beyond the repository owner (nndl) and referenced materials in the README.

Caveats

• License: MIT
• Created: 2017-03-29
• Last push: 2026-06-23
• Issues: 0 open issues
• Language: Python
• Primary topics: agent, fudannlp, llm, openmoss, step-by-step