GPT-4o详细中文注释的Colab
中文注释链接:https://github.com/Czi24/Awesome-MLLM-LLM-Colab/tree/master/Courses/Prompt-Compression-and-Query-Optimization
观看视频
1 浏览器下载插件
沉浸式翻译
设置你需要用的翻译软件
2 打开官方视频
视频官方地址:https://learn.deeplearning.ai/courses/prompt-compression-and-query-optimization/lesson/1/introduction
打开自动开启双语字幕
Build a Large Language Model (From Scratch)
如果这个仓库对你有帮助,请点个star,并私聊我,我将发送给你《Build a Large Language Model (From Scratch)》的电子版,包括GPT-4翻译的全书PDF,方便你做笔记。
仓库:https://github.com/Czi24/Awesome-MLLM-LLM-Colab
课程1:Prompt Compression and Query Optimization
Prompt Compression and Query Optimization |
提示压缩与查询优化 |
Introduction |
介绍 |
Vanilla Vector Search |
基础向量搜索 |
Filtering With Metadata |
元数据过滤 |
Projections |
投影 |
Boosting |
提升 |
Prompt Compression |
提示压缩 |
Conclusion |
结论 |
Appendix-Tips and Help |
附录-提示和帮助 |
课程2:Carbon Aware Computing for GenAI developers
Carbon Aware Computing for GenAI developers |
面向生成式AI开发人员的碳感知计算 |
Introduction |
介绍 |
The Carbon Footprint of Machine Learning |
机器学习的碳足迹 |
Exploring Carbon Intensity on the Grid |
探索电网中的碳强度 |
Training Models in Low Carbon Regions |
在低碳地区训练模型 |
Using Real-Time Energy Data for Low-Carbon Training |
使用实时能源数据进行低碳训练 |
Understanding your Google Cloud Footprint |
了解你的谷歌云碳足迹 |
Next steps |
下一步 |
Conclusion |
结论 |
Google Cloud Setup |
谷歌云设置 |
课程3:Function-calling and data extraction with LLMs
Function-calling and data extraction with LLMs |
使用LLMs进行函数调用和数据提取 |
Introduction |
介绍 |
What is function calling |
什么是函数调用 |
Function calling variations |
函数调用的变体 |
Interfacing with external tools |
与外部工具的接口 |
Structured Extraction |
结构化提取 |
Applications |
应用 |
Course project dialog processing |
课程项目对话处理 |
Conclusion |
结论 |
课程4:Building Your Own Database Agent
Building Your Own Database Agent |
构建你自己的数据库代理 |
Introduction |
介绍 |
Your First AI Agent |
你的第一个AI代理 |
Interacting with a CSV Data |
处理CSV数据 |
Connecting to a SQL Database |
连接SQL数据库 |
Azure OpenAI Function Calling Feature |
Azure OpenAI函数调用功能 |
Leveraging Assistants API for SQL Databases |
利用助手API处理SQL数据库 |
Conclusion |
结论 |
课程5:AI Agents in LangGraph
AI Agents in LangGraph |
LangGraph中的AI代理 |
Introduction |
介绍 |
Build an Agent from Scratch |
从头构建代理 |
LangGraph Components |
LangGraph组件 |
Agentic Search Tools |
代理搜索工具 |
Persistence and Streaming |
持久性与流媒体 |
Human in the loop |
人在回路中 |
Essay Writer |
文章写作 |
LangChain Resources |
LangChain资源 |
Conclusion |
结论 |
课程6:AI Agentic Design Patterns with AutoGen
AI Agentic Design Patterns with AutoGen |
使用AutoGen的AI代理设计模式 |
Introduction |
介绍 |
Multi-Agent Conversation and Stand-up Comedy |
多代理对话与单口喜剧 |
Sequential Chats and Customer Onboarding |
连续聊天与客户入职 |
Reflection and Blogpost Writing |
反思与博客写作 |
Tool Use and Conversational Chess |
工具使用与对话象棋 |
Coding and Financial Analysis |
编码与财务分析 |
Planning and Stock Report Generation |
规划与股票报告生成 |
Conclusion |
结论 |
课程7:Introduction to on-device AI
Introduction to on-device AI |
设备端AI简介 |
Introduction |
介绍 |
Why on-device |
为什么选择设备端AI |
Deploying Segmentation Models On-Device |
部署设备端分割模型 |
Preparing for on-device deployment |
准备设备端部署 |
Quantizing Models |
量化模型 |
Device Integration |
设备集成 |
Conclusion |
结论 |
Appendix - Building the App |
附录 - 构建应用 |
Appendix - Tips and Help |
附录 - 提示和帮助 |
课程8:Multi AI Agent Systems with crewAI
Multi AI Agent Systems with crewAI |
使用crewAI的多AI代理系统 |
Introduction |
介绍 |
Overview |
概览 |
AI Agents |
AI代理 |
Create agents to research and write an article (code) |
创建代理进行研究和写文章(代码) |
Key elements of AI agents |
AI代理的关键要素 |
Multi agent customer support automation (code) |
多代理客户支持自动化(代码) |
Mental framework for agent creation |
代理创建的思维框架 |
Key elements of agent tools |
代理工具的关键要素 |
Tools for a customer outreach campaign (code) |
客户外展活动的工具(代码) |
Recap of tools |
工具回顾 |
Key elements of well defined tasks |
定义明确任务的关键要素 |
Automate event planning (code) |
自动化事件规划(代码) |
Recap on tasks |
任务回顾 |
Multi agent collaboration |
多代理协作 |
Multi agent collaboration for financial analysis (code) |
多代理财务分析协作(代码) |
Build a crew to tailor job applications (code) |
创建团队定制工作申请(代码) |
Next steps with AI agent systems |
AI代理系统的下一步 |
Conclusion |
结论 |
How to get your completion badge |
如何获得完成徽章 |
课程9:Building Multimodal Search and RAG
Building Multimodal Search and RAG |
构建多模态搜索和RAG |
Introduction |
介绍 |
Overview of Multimodality |
多模态概述 |
Multimodal Search |
多模态搜索 |
Large Multimodal Models (LMMs) |
大型多模态模型(LMMs) |
Multimodal RAG (MM-RAG) |
多模态RAG(MM-RAG) |
Industry Applications |
行业应用 |
Multimodal Recommender System |
多模态推荐系统 |
Conclusion |
结论 |
Appendix - Tips and Help |
附录 - 提示和帮助 |
课程10:Building Agentic RAG with Llamaindex
Building Agentic RAG with Llamaindex |
使用Llamaindex构建Agentic RAG |
Introduction |
介绍 |
Router Query Engine |
路由查询引擎 |
Tool Calling |
工具调用 |
Building an Agent Reasoning Loop |
构建代理推理循环 |
Building a Multi-Document Agent |
构建多文档代理 |
Conclusion |
结论 |
课程11:Quantization in Depth
Quantization in Depth |
深入量化 |
Introduction |
介绍 |
Overview |
概览 |
Quantize and De-quantize a Tensor |
量化和反量化张量 |
Get the Scale and Zero Point |
获取比例和零点 |
Symmetric vs Asymmetric Mode |
对称模式与非对称模式 |
Finer Granularity for more Precision |
更精细的粒度以提高精度 |
Per Channel Quantization |
每通道量化 |
Per Group Quantization |
每组量化 |
Quantizing Weights & Activations for Inference |
推理的权重和激活量化 |
Custom Build an 8-Bit Quantizer |
自定义构建8位量化器 |
Replace PyTorch layers with Quantized Layers |
用量化层替换PyTorch层 |
Quantize any Open Source PyTorch Model |
量化任何开源PyTorch模型 |
Load your Quantized Weights from HuggingFace Hub |
从HuggingFace Hub加载量化权重 |
Weights Packing |
权重打包 |
Packing 2-bit Weights |
打包2位权重 |
Unpacking 2-Bit Weights |
解包2位权重 |
Beyond Linear Quantization |
超越线性量化 |
Conclusion |
结论 |
课程12:Prompt Engineering for Vision Models
Prompt Engineering for Vision Models |
视觉模型的提示工程 |
Introduction |
介绍 |
Overview |
概览 |
Image Segmentation |
图像分割 |
Object Detection |
目标检测 |
Image Generation |
图像生成 |
Fine-tuning |
微调 |
Conclusion |
结论 |
Appendix |
附录 |
课程13:Getting Started with Mistral
Getting Started with Mistral |
入门Mistral |
Introduction |
介绍 |
Overview |
概览 |
Prompting |
提示 |
Model Selection |
模型选择 |
Function Calling |
函数调用 |
RAG from Scratch |
从零开始构建RAG |
Chatbot |
聊天机器人 |
Conclusion |
结论 |
课程14:Quantization Fundamentals with Hugging Face
Quantization Fundamentals with Hugging Face |
Hugging Face的量化基础 |
Introduction |
介绍 |
Handling Big Models |
处理大模型 |
Data Types and Sizes |
数据类型和大小 |
Loading Models by data type |
按数据类型加载模型 |
Quantization Theory |
量化理论 |
Quantization of LLMs |
LLMs的量化 |
Conclusion |
结论 |
课程15:Preprocessing Unstructured Data for LLM Applications
Preprocessing Unstructured Data for LLM Applications |
预处理LLM应用程序的非结构化数据 |
Introduction |
介绍 |
Overview of LLM Data Preprocessing |
LLM数据预处理概述 |
Normalizing the Content |
内容规范化 |
Metadata Extraction and Chunking |
元数据提取和分块 |
Preprocessing PDFs and Images |
预处理PDF和图像 |
Extracting Tables |
提取表格 |
Build Your Own RAG Bot |
构建你自己的RAG机器人 |
Conclusion |
结论 |
Appendix - Tips and Help |
附录 - 提示和帮助 |
课程16:Red Teaming LLM Applications
Red Teaming LLM Applications |
LLM应用程序的红队测试 |
Introduction |
介绍 |
Overview of LLM Vulnerabilities |
LLM漏洞概述 |
Red Teaming LLMs |
红队测试LLMs |
Red Teaming at Scale |
大规模红队测试 |
Red Teaming LLMs with LLMs |
用LLMs进行红队测试 |
A Full Red Teaming Assessment |
全面的红队评估 |
Conclusion |
结论 |
课程17:JavaScript RAG Web Apps with LlamaIndex
JavaScript RAG Web Apps with LlamaIndex |
使用LlamaIndex的JavaScript RAG Web应用 |
Introduction |
介绍 |
Getting started with RAG |
入门RAG |
Build a full-stack web app |
构建全栈Web应用 |
Advanced queries with Agents |
使用代理的高级查询 |
Production-ready techniques |
生产就绪技术 |
Conclusion |
结论 |
课程18:Efficiently Serving LLMs
Efficiently Serving LLMs |
高效服务LLMs |
Introduction |
介绍 |
Text Generation |
文本生成 |
Batching |
批处理 |
Continuous Batching |
连续批处理 |
Quantization |
量化 |
Low-Rank Adaptation |
低秩适应 |
Multi-LoRA inference |
多LoRA推理 |
LoRAX |
LoRAX |
Conclusion |
结论 |
课程19:Knowledge Graphs for RAG
Knowledge Graphs for RAG |
RAG的知识图谱 |
Introduction |
介绍 |
Knowledge Graph Fundamentals |
知识图谱基础 |
Querying Knowledge Graphs |
查询知识图谱 |
Preparing Text for RAG |
为RAG准备文本 |
Constructing a Knowledge Graph from Text Documents |
从文本文件构建知识图谱 |
Adding Relationships to the SEC Knowledge Graph |
向SEC知识图谱添加关系 |
Expanding the SEC Knowledge Graph |
扩展SEC知识图谱 |
Chatting with the Knowledge Graph |
与知识图谱聊天 |
Conclusion |
结论 |
课程20:Open Source Models with Hugging Face
Open Source Models with Hugging Face |
使用Hugging Face的开源模型 |
Introduction |
介绍 |
Selecting models |
选择模型 |
Natural Language Processing (NLP) |
自然语言处理(NLP) |
Translation and Summarization |
翻译和摘要 |
Sentence Embeddings |
句子嵌入 |
Zero-Shot Audio Classification |
零样本音频分类 |
Automatic Speech Recognition |
自动语音识别 |
Text to Speech |
文本转语音 |
Object Detection |
目标检测 |
Image Segmentation |
图像分割 |
Image Retrieval |
图像检索 |
Image Captioning |
图像标题生成 |
Multimodal Visual Question Answering |
多模态视觉问答 |
Zero-Shot Image Classification |
零样本图像分类 |
Deployment |
部署 |
Conclusion |
结论 |
课程21:Prompt Engineering with Llama 2&3
Prompt Engineering with Llama 2&3 |
使用Llama 2&3进行提示工程 |
Introduction |
介绍 |
Overview of Llama Models |
Llama模型概述 |
Getting Started with Llama 2 & 3 |
Llama 2&3入门 |
Multi-turn Conversations |
多轮对话 |
Prompt Engineering Techniques |
提示工程技术 |
Comparing Different Llama 2 & 3 models |
比较不同的Llama 2&3模型 |
Code Llama |
代码Llama |
Llama Guard |
Llama卫士 |
Walkthrough of Llama Helper Function (Optional) |
Llama助手函数演练(可选) |
Conclusion |
结论 |
课程22:Serverless LLM Apps Amazon Bedrock
Serverless LLM Apps Amazon Bedrock |
使用Amazon Bedrock的无服务器LLM应用 |
Introduction |
介绍 |
Your first generations with Amazon Bedrock |
使用Amazon Bedrock生成第一个结果 |
Summarize an audio file |
总结音频文件 |
Enable logging |
启用日志记录 |
Deploy an AWS Lambda function |
部署AWS Lambda函数 |
Event-driven generation |
事件驱动生成 |
Conclusion |
结论 |
课程23:Building Applications with Vector Databases
Building Applications with Vector Databases |
使用向量数据库构建应用 |
Introduction |
介绍 |
Semantic Search |
语义搜索 |
Retrieval Augmented Generation (RAG) |
检索增强生成(RAG) |
Recommender Systems |
推荐系统 |
Hybrid Search |
混合搜索 |
Facial Similarity Search |
面部相似性搜索 |
Anomaly Detection |
异常检测 |
Conclusion |
结论 |
课程24:Automated Testing for LLMOps
Automated Testing for LLMOps |
LLMOps的自动化测试 |
Introduction |
介绍 |
Introduction to Continuous Integration (CI) |
持续集成(CI)介绍 |
Overview of Automated Evals |
自动评估概述 |
Automating Model-Graded Evals |
自动化模型评分评估 |
Comprehensive Testing Framework |
综合测试框架 |
Conclusion |
结论 |
Optional: Exploring the CircleCI config file |
可选:探索CircleCI配置文件 |
课程25:LLMOps
LLMOps |
LLMOps |
Introduction |
介绍 |
The Fundamentals |
基础知识 |
Data Preparation |
数据准备 |
Automation and Orchestration with Pipelines |
流水线的自动化和编排 |
Prediction, Prompts, Safety |
预测、提示、安全 |
Conclusion |
结论 |
Next Step |
下一步 |
课程26:Build LLM Apps with LangChain.js
Build LLM Apps with LangChain.js |
使用LangChain.js构建LLM应用程序 |
Introduction |
介绍 |
Building Blocks |
构建模块 |
Loading and preparing data |
加载和准备数据 |
Vectorstores and embeddings |
向量存储和嵌入 |
Question answering |
问答 |
Conversational question answering |
对话问答 |
Shipping as a web API |
作为Web API发布 |
Conclusion |
结论 |
Next Step |
下一步 |
课程27:Advanced Retrieval for AI with Chroma
Advanced Retrieval for AI with Chroma |
使用Chroma进行高级检索 |
Introduction |
介绍 |
Overview of embeddings-based retrieval |
基于嵌入的检索概述 |
Pitfalls of retrieval - when simple vector search fails |
检索的陷阱 - 当简单向量搜索失败时 |
Query Expansion |
查询扩展 |
Cross-encoder re-ranking |
交叉编码器重新排序 |
Embedding adaptors |
嵌入适配器 |
Other Techniques |
其他技术 |
课程28:Reinforcement Learning From Human Feedback
Reinforcement Learning From Human Feedback |
从人类反馈中进行强化学习 |
Introduction |
介绍 |
How does RLHF work |
RLHF如何工作 |
Datasets for RL training |
强化学习的数据集 |
Tune an LLM with RLHF |
使用RLHF调整LLM |
Evaluate the tuned model |
评估调整后的模型 |
Google Cloud Setup |
Google Cloud设置 |
Conclusion |
结论 |
课程29:Building and Evaluating Advanced RAG
Building and Evaluating Advanced RAG |
构建和评估高级RAG |
Introduction |
介绍 |
Advanced RAG Pipeline |
高级RAG流水线 |
RAG Triad of metrics |
RAG的三重指标 |
Sentence-window retrieval |
句子窗口检索 |
Auto-merging retrieval |
自动合并检索 |
Conclusion |
结论 |
课程30:Quality and Safety for LLM Applications
Quality and Safety for LLM Applications |
LLM应用的质量和安全 |
Introduction |
介绍 |
Overview |
概览 |
Hallucinations |
幻觉 |
Data Leakage |
数据泄露 |
Refusals and prompt injections |
拒绝和提示注入 |
Passive and active monitoring |
被动和主动监控 |
Conclusion |
结论 |
课程31:Vector Databases: from Embeddings to Applications
Vector Databases: from Embeddings to Applications |
向量数据库:从嵌入到应用 |
Introduction |
介绍 |
How to Obtain Vector Representations of Data |
如何获取数据的向量表示 |
Search for Similar Vectors |
搜索相似向量 |
Approximate nearest neighbours |
近似最近邻 |
Vector Databases |
向量数据库 |
Sparse, Dense, and Hybrid Search |
稀疏、密集和混合搜索 |
Application - Multilingual Search |
应用 - 多语言搜索 |
Conclusion |
结论 |
课程32:Functions, Tools and Agents with LangChain
Functions, Tools and Agents with LangChain |
使用LangChain的函数、工具和代理 |
Introduction |
介绍 |
OpenAI Function Calling |
OpenAI函数调用 |
LangChain Expression Language (LCEL) |
LangChain表达语言(LCEL) |
OpenAI Function Calling in LangChain |
在LangChain中调用OpenAI函数 |
Tagging and Extraction |
标记和提取 |
Tools and Routing |
工具和路由 |
Conversational Agent |
会话代理 |
Conclusion |
结论 |
课程33:Pair Programming with a Large Language Model
Pair Programming with a Large Language Model |
使用大型语言模型进行结对编程 |
Introduction |
介绍 |
Getting Started |
入门 |
Using a String Template |
使用字符串模板 |
Pair Programming Scenarios |
结对编程场景 |
Technical Debt |
技术债务 |
Conclusion |
结论 |
课程34:Understanding and Applying Text Embeddings
Understanding and Applying Text Embeddings |
理解和应用文本嵌入 |
Introduction |
介绍 |
Getting Started With Text Embeddings |
文本嵌入入门 |
Understanding Text Embeddings |
理解文本嵌入 |
Visualizing Embeddings |
可视化嵌入 |
Applications of Embeddings |
嵌入的应用 |
Text Generation with Vertex AI |
使用Vertex AI生成文本 |
Building a Q&A System Using Semantic Search |
使用语义搜索构建问答系统 |
Optional - Google Cloud Setup |
可选 - Google Cloud设置 |
Conclusion |
结论 |
课程35:How Business Thinkers Can Start Building AI Plugins With Semantic Kernel
How Business Thinkers Can Start Building AI Plugins With Semantic Kernel |
商业思考者如何使用语义内核开始构建AI插件 |
Introduction |
介绍 |
Semantic Kernel is Like Your AI Cooking Kitchen |
语义内核就像你的AI烹饪厨房 |
Cooking Up Flavorful SWOTs with the Kernel |
用内核做出美味的SWOT分析 |
Organizing The Tools You Make for Later Reuse |
组织你制作的工具以备后用 |
Frozen Dinner The Design Thinking Meal |
冷冻晚餐的设计思维餐 |
Dont Forget to Save the Generated Dripping or The Gravy |
不要忘记保存生成的油滴或肉汁 |
A Kitchen That Responds to Your I’m Hungry is More Than Feasible |
响应你的“我饿了”的厨房是完全可行的 |
There’s a Fully-Outfitted Professional-Grade Kitchen Ready For You |
有一个装备齐全的专业厨房为你准备好了 |
Conclusion |
结论 |
课程36:Finetuning Large Language Models
Finetuning Large Language Models |
微调大型语言模型 |
Introduction |
介绍 |
Why finetune |
为什么微调 |
Where finetuning fits in |
微调的适用场景 |
Instruction finetuning |
指令微调 |
Data preparation |
数据准备 |
Training process |
训练过程 |
Evaluation and iteration |
评估和迭代 |
Consideration on getting started now |
现在开始的考虑因素 |
Conclusion |
结论 |
课程37:Large Language Models with Semantic Search
Large Language Models with Semantic Search |
具有语义搜索的大型语言模型 |
Introduction |
介绍 |
Keyword Search |
关键词搜索 |
Embeddings |
嵌入 |
Dense Retrieval |
稠密检索 |
ReRank |
重新排序 |
Generating Answers |
生成答案 |
Conclusion |
结论 |
课程38:Evaluating and Debugging Generative AI
Evaluating and Debugging Generative AI |
评估和调试生成式AI |
Introduction |
介绍 |
Instrument W&B |
工具W&B |
Training a Diffusion Model with W&B |
使用W&B训练扩散模型 |
Evaluating Diffusion Models |
评估扩散模型 |
LLM Evaluation and Tracing with W&B |
使用W&B进行LLM评估和追踪 |
Finetuning a language model |
微调语言模型 |
Conclusion |
结论 |
课程39:Building Generative AI Applications with Gradio
Building Generative AI Applications with Gradio |
使用Gradio构建生成式AI应用 |
Introduction |
介绍 |
NLP Tasks interface |
NLP任务界面 |
Image Captioning app |
图像字幕应用 |
Image generation app |
图像生成应用 |
Describe and Generate Game |
描述和生成游戏 |
Chat with any LLM |
与任何LLM聊天 |
Conclusion |
结论 |
课程40:LangChain Chat with Your Data
LangChain Chat with Your Data |
使用LangChain与数据聊天 |
Introduction |
介绍 |
Document Loading |
文档加载 |
Document Splitting |
文档拆分 |
Vectorstores and Embedding |
向量存储和嵌入 |
Retrieval |
检索 |
Question Answering |
问答 |
Chat |
聊天 |
Conclusion |
结论 |
课程41:Building Systems with the ****** API
Building Systems with the ****** API |
使用****** API构建系统 |
Introduction |
介绍 |
Language Models, the Chat Format and Tokens |
语言模型、聊天格式和词元 |
Classification |
分类 |
Moderation |
审核 |
Chain of Thought Reasoning |
思维链推理 |
Chaining Prompts |
链接提示 |
Check Outputs |
检查输出 |
Evaluation |
评估 |
Evaluation Part I |
评估第一部分 |
Evaluation Part II |
评估第二部分 |
Summary |
总结 |
课程42:How Diffusion Models Work
How Diffusion Models Work |
扩散模型的工作原理 |
Introduction |
介绍 |
Intuition |
直觉 |
Sampling |
采样 |
Neural Network |
神经网络 |
Training |
训练 |
Controlling |
控制 |
Speeding Up |
加速 |
Summary |
总结 |