How alibabaclouddocs/aliyun_acp_learning works
Project type and purpose
• This is an educational “project‐style” repository that walks learners through building, evaluating, tuning and deploying a Retrieval‑Augmented Generation (RAG)‑based question‑answering chatbot using large language models (LLMs).
• Its primary “production” system is the P2 chatbot: code modules and notebooks that implement data ingestion, prompt engineering, retrieval (via Llama‑Index), LLM inference, security checks, evaluation, fine‑tuning and deployment.
File structure
• Top level:
• config/load_key.py – centralized loader for API keys
• p1_课程准备 – environment setup notebooks
• p2_构造大模型问答系统 – the core chatbot project, with:
– chatbot/ (evaluate.py, llm.py, rag.py)
– config/load_key.py
– docs/ (PDFs/DOCs used as knowledge base)
– ragas_prompt/ (prompt templates)
– resources/ (JSONL data for testing, training, RAG slices)
– utils/ (custom LLM adapters, evaluation helpers, security sub‑packages: audio, image, text, kb_access_control, OSS service, security manager)
• p3_总结与展望 – wrap‑up notebooks
• Other: requirements.txt for dependencies; README.md for course description.
Extract architectural insights from README
• Technologies: Python, OpenAI (or Alicloud’s Baichuan), LangChain/LlamaIndex, JSONL datasets, RAGAS evaluation metrics, vLLM for deployment, Alibaba Cloud function/ECS/PAI.
• Workflow: build → optimize prompts → automate evaluation → extend via plugins → apply RAG → fine‑tune → deploy.
Final advice
• Explicitly how data moves from raw documents → retrieval → LLM → evaluation → deployment.