适用人群
This bootcamp is suitable for students who are interested in the application of large language models. It can help students improve their project background and enhance their competitiveness for future study and career development in related fields.
你将收获
熟练掌握如何使用LLM和相关工具来搭建各种复杂的应用
完成6个LLM实战项目
项目对于DS / AI / MLE 方向的留学申请/求职会有相关背景的大幅度提升
你将学习的技能
使用RAG模式来实现复杂的LLM应用场景
通过构建AI Agent 来实现通用人工智能 (AGI)
课题介绍
In the era of artificial intelligence and large models, large language models (LLM) have become an indispensable tool, providing powerful support for various real-life application scenarios. From simple text classification to complex logical inference and code generation, LLM is changing the way we deal with problems. LLM can do much more than simple chat conversations like ChatGPT, and people are constantly exploring the potential of LLM. How to apply LLM to complex application scenarios and how to build an application based on LLM have become increasingly important skills. This project will focus on LLM applications and provide students with practical LLM project experience through a large number of practical tasks.
课程大纲
The bootcamp content mainly includes: introduction to LLM, completing simple tasks through Prompt Engineering, using RAG mode to realize complex LLM application scenarios, and realizing general artificial intelligence (AGI) by building AI Agent
1
LLM 入门：什么是LLM / GPT； LLM 应用场景介绍：Q&A, Chatbot, Agent, etc. LLM 的核心模型：Transformer & Self-Attention; BERT vs GPT
考察：代码练习
授课2小时; 学习3小时
2
通过指令让LLM完成指定任务 Prompt Engineering；In-context learning； Reasoning 实战练习1：Text classification using LLM
考察：代码练习
授课2小时; 学习3小时
3
搭建基于LLM的应用 生成/储存/查找 embedding：embedding model, VectorDB and Similarity Search 常用工具介绍：Langchain / PromptFlow RAG框架介绍，搭建基于RAG的应用 实战练习2: Text Embedding & Retrieval for large document 实战练习3: Build a chatbot using Langchain 实战练习4：Build a RAG-based Q&A system
考察：代码练习
授课2小时; 学习3小时
4
定制/微调LLM LLM模型的训练过程：Pre-training, Finetuning, Alignment LLM中重要的数学计算和实现代码 微调模型的技巧: PEFT, KV Cache, Fast attention, ... 实战练习5: Implement a transformer decoder block
考察：代码练习
授课3小时; 学习2小时
5
实战项目：Customize LLM for Q&A system 数据准备：把原始数据处理成大模型需要的形式 评估计算资源：根据模型大小和数据量计算需要的GPU和训练时间 训练模型：在预训练的LLM上使用LoRA进行微调 评估模型：评估模型生成文本的质量 部署和搭建应用：使用微调好的模型搭建一个Q&A system
考察：代码练习
授课3小时; 学习3小时
课程导师
Joey
经历
甲骨文数据科学家（美国）惠普机器学习工程师（美国）
教育
宾夕法尼亚大学数据科学硕士
