介绍
获取前 100+ 个精选的 LLM 面试问题，了解如何准备生成式 AI 或 LLM 面试准备和大型语言模型 （LLM） 面试准备的学习路径。
This article explains learning path for large language models (LLMs) interview preparation. You will find below details in this article:
本文介绍了大型语言模型 （LLM） 面试准备的学习路径。您将在本文中找到以下详细信息：
- Road map路线图
- Prompt engineering & basics of LLM提示工程和LLM基础知识
- Retrieval augmented generation (RAG)检索增强生成 （RAG）
- Chunking strategies分块策略
- Embedding Models嵌入模型
- Internal working of vector DB矢量数据库的内部工作
- Advanced search algorithms高级搜索算法
- Language models internal working语言模型内部工作
- Supervised fine-tuning of LLMLLM 的监督微调
- Preference Alignment (RLHF/DPO)
首选项对齐 （RLHF/DPO） - Evaluation of LLM systemLLM 系统的评估
- Hallucination control techniques幻觉控制技术
- Deployment of LLM部署 LLM
- Agent-based system基于代理的系统
- Prompt Hacking提示黑客攻击
- Case Study & Scenario-based question案例研究和基于情景的问题
Roadmap路线图
Prompt engineering & basics of LLM提示工程和LLM基础知识
- Question 1 : What is the difference between Predictive/ Discriminative AI and generative AI?
问题 1：预测性/判别性 AI 和生成式 AI 有什么区别？ - Question 2: What is LLM & how LLMs are trained?
问题2：什么是LLM&LLM是如何被训练的？ - Question 3: What is a token in the language model?
问题 3：语言模型中的令牌是什么？ - Question 4: How to estimate the cost of running a SaaS-based & Open source LLM model?
问题4：如何估算运行基于SaaS和开源的LLM模型的成本？ - Question 5: Explain the Temperature parameter and how to set it.
问题 5：解释 Temperature 参数以及如何设置它。 - Question 6: What are different decoding strategies for picking output tokens?
问题 6：选择输出 Token 有哪些不同的解码策略？ - Question 7: What are the different ways you can define stopping criteria in a large language model?
问题 7：在大型语言模型中定义停止条件的方法有哪些？ - Question 8: How to use stop sequence in LLMs?
问题 8：如何在 LLM 中使用停止序列？ - Question 9: Explain the basic structure of prompt engineering.
问题 9：解释提示工程的基本结构。 - Question 10: Explain the type of prompt engineering
问题 10：解释提示工程的类型 - **Question 11: **Explain In-Context Learning问题 11：解释情境学习
- Question 12: What are some of the aspects to keep in mind while using few-shots prompting?
问题 12：使用小镜头提示时需要记住哪些方面？ - Question 13: What are certain strategies to write good prompts?
问题 13：写好提示的策略是什么？ - Question 14: What is hallucination & how can it be controlled using prompt engineering?
问题14：什么是幻觉，如何通过提示工程来控制它？ - Question 15: How do I improve the reasoning ability of my LLM through prompt engineering?
问题 15：如何通过提示工程提高我的 LLM 的推理能力？ - Question 16: How to improve LLM reasoning if your COT prompt fails?
问题 16：如果您的 COT 提示失败，如何改进 LLM 推理？
Want to find out correct and accurate answers? Look for our LLM Interview Course
想找出正确和准确的答案吗？寻找我们的 LLM 面试课程
- 100+ Questions spanning 14 categories
100+ 问题，跨越 14 个类别 - Curated 100+ assessments for each category
为每个类别策划 100+ 评估 - Well-researched real-world interview questions based on FAANG & Fortune 500 companies
基于FAANG和财富500强公司的真实面试问题 - Focus on Visual learning专注于视觉学习
- Real Case Studies & Certification真实案例研究和认证
50% off Coupon Code --- LLM50
优惠券代码 50% 折扣 --- LLM50
Coupon is valid till 30th May 2024
优惠券有效期至 2024 年 5 月 30 日
Link for the course ---课程链接 ---
大型语言模型（LLM）面试问答课程
通过### 这个全面的大型语言模型（LLM）面试问答课程，深入探索AI的世界...
Retrieval augmented generation (RAG)检索增强生成 （RAG）
- Question 1: How to increase accuracy, and reliability & make answers verifiable in LLM?
问题1：如何在LLM中提高准确性和可靠性并使答案可验证？ - **Question 2: **How does Retrieval augmented generation (RAG) work?
问题 2：检索增强生成 （RAG） 的工作原理是什么？ - Question 3: What are some of the benefits of using the RAG system?
问题 3：使用 RAG 系统有哪些好处？ - Question 4: What are the architecture patterns you see when you want to customize your LLM with proprietary data?
问题 4：当您想使用专有数据自定义 LLM 时，您会看到哪些架构模式？ - Question 5: When should I use Fine-tuning instead of RAG?
问题 5：我什么时候应该使用 Fine-tuning 而不是 RAG？
Chunking strategies分块策略
- Question 1: What is chunking and why do we chunk our data?
问题 1：什么是分块，为什么我们要对数据进行分块？ - Question 2: What are factors influences chunk size?
问题 2：影响块大小的因素有哪些？ - Question 3: What are the different types of chunking methods available?
问题 3：有哪些不同类型的分块方法可用？ - Question 4: How to find the ideal chunk size?
问题 4：如何找到理想的块大小？
Embedding Models嵌入模型
- Question 1: What are vector embeddings? And what is an embedding model?
问题 1：什么是向量嵌入？什么是嵌入模型？ - Question 2: How embedding model is used in the context of LLM application?
问题 2：在 LLM 应用程序的上下文中如何使用嵌入模型？ - Question 3: What is the difference between embedding short and long content?
问题 3：嵌入短内容和长内容有什么区别？ - Question 4: How to benchmark embedding models on your data?
问题 4：如何对数据进行嵌入模型基准测试？ - Question 5: Walk me through the steps of improving the sentence transformer model used for embedding
问题 5：请向我介绍改进用于嵌入的句子转换模型的步骤
Internal working of vector DB矢量数据库的内部工作
- Question 1: What is vector DB?
问题 1：什么是 vector DB？ - Question 2: How vector DB is different from traditional databases?
问题 2：矢量数据库与传统数据库有何不同？ - Question 3: How does a vector database work?
问题 3：矢量数据库如何工作？ - Question 4: Explain the difference between vector index, vector DB & vector plugins.
问题4：解释向量索引、向量DB和向量插件之间的区别。 - Question 5: What are different vector search strategies?
问题 5：有哪些不同的向量搜索策略？ - Question 6: How does clustering reduce search space? When does it fail and how can we mitigate these failures?
问题 6：聚类如何减少搜索空间？它何时会失败，我们如何减轻这些失败？ - Question 7: Explain the Random projection index.
问题 7：解释随机投影指数。 - Question 8: Explain the Localitysensitive hashing (LHS) indexing method?
问题 8：解释 Localitysensitive 哈希 （LHS） 索引方法？ - Question 9: Explain the product quantization (PQ) indexing method
问题 9：解释乘积量化 （PQ） 标定方法 - Question 10: Compare different Vector indexes and given a scenario, which vector index you would use for a project?
问题 10：比较不同的 Vector 索引，并在给定一个场景中，您将为项目使用哪个 Vector 索引？ - Question 11: How would you decide on ideal search similarity metrics for the use case?
问题 11：您将如何为用例确定理想的搜索相似度指标？ - Question 12: Explain the different types and challenges associated with filtering in vector DB.
问题 12：解释与矢量 DB 中的过滤相关的不同类型和挑战。 - Question 13: How do you determine the best vector database for your needs?
问题 13：您如何确定最适合您需求的向量数据库？
Advanced search algorithms高级搜索算法
- Question 1: Why it's important to have very good search
问题 1：为什么拥有非常好的搜索很重要 - Question 2: What are the architecture patterns for information retrieval & semantic search, and their use cases?
问题2：信息检索和语义搜索的架构模式是什么，以及它们的用例是什么？ - Question 3: How can you achieve efficient and accurate search results in large scale datasets?
问题 3：如何在大规模数据集中获得高效准确的搜索结果？ - Question 4: Explain the keyword-based retrieval method
问题 4：解释基于关键字的检索方法 - Question 5: How to fine-tune re-ranking models?
问题 5：如何微调重新排名模型？ - Question 6: Explain most common metric used in information retrieval and when it fails?
问题 6：解释信息检索中最常用的指标以及何时失败？ - Question 7: I have a recommendation system, which metric should I use to evaluate the system?
问题 7：我有一个推荐系统，我应该使用哪个指标来评估该系统？ - Question 8: Compare different information retrieval metrics and which one to use when?
问题 8：比较不同的信息检索指标以及何时使用哪一个？
Language models internal working语言模型内部工作
- Question 1: Detailed understanding of the concept of selfattention
问题 1：详细了解自我注意的概念 - Question 2: Overcoming the disadvantages of the self-attention mechanism
问题 2：克服自我注意机制的缺点 - Question 3: Understanding positional encoding问题 3：了解位置编码
- Question 4: Detailed explanation of Transformer architecture
问题 4：Transformer 架构详解 - Question 5: Advantages of using a transformer instead of LSTM.
问题 5：使用变压器代替 LSTM 的优势。 - Question 6: Difference between local attention and global attention
问题 6：本地关注度与全球关注度的差异 - Question 7: Understanding the computational and memory demands of transformers
问题 7：了解 transformer 的计算和内存需求 - Question 8: Increasing the context length of an LLM.
问题 8：增加 LLM 的上下文长度。 - Question 9: How to Optimizing transformer architecture for large vocabularies
问题 9：如何优化大词汇表的 transformer 架构 - Question 10: What is a mixture of expert models?
问题 10：什么是专家模型的混合？
Supervised finetuning of LLMLLM 的监督微调
- **Question 1: **What is finetuning and why it' s needed in LLM?
问题 1：什么是微调，为什么在 LLM 中需要它？ - **Question 2: **Which scenario do we need to finetune LLM?
问题 2：我们需要在哪种情况下微调 LLM？ - **Question 3: **How to make the decision of finetuning?
问题 3：如何做出微调的决定？ - Question 4: How do you create a fine-tuning dataset for Q&A?
问题 4：如何为 Q&A 创建微调数据集？ - Question 5: How do you improve the model to answer only if there is sufficient context for doing so?
问题 5：你如何改进模型，只有在有足够的上下文的情况下才能回答？ - Question 6: How to set hyperparameter for fine-tuning
问题 6：如何设置超参数进行微调 - Question 7: How to estimate infra requirements for fine-tuning LLM?
问题 7：如何估计微调 LLM 的基础设施需求？ - Question 8: How do you finetune LLM on consumer hardware?
问题 8：如何在消费类硬件上微调 LLM？ - Question 9: What are the different categories of the PEFT method?
问题 9：PEFT 方法有哪些不同的类别？ - Question 10: Explain different reparameterized methods for finetuning LLM?
问题 10：解释微调 LLM 的不同重新参数化方法？ - Question 11: What is catastrophic forgetting in the context of LLMs?
问题 11：在 LLM 的背景下，什么是灾难性遗忘？
Preference Alignment (RLHF/DPO)
首选项对齐 （RLHF/DPO）
- **Question 1: **At which stage you will decide to go for the Preference alignment type of method rather than SFT?
问题 1：在哪个阶段，您将决定使用 Preference alignment 类型的方法而不是 SFT？ - Question 2: Explain Different Preference Alignment Methods?
问题 2：解释不同的首选项对齐方法？ - Question 3: What is RLHF, and how is it used?
问题 3：什么是 RLHF，如何使用它？ - Question 4: Explain the reward hacking issue in RLHF.
问题 4：解释 RLHF 中的奖励黑客问题。
Evaluation of LLM systemLLM 系统的评估
- Question 1: How do you evaluate the best LLM model for your use case?
问题 1：您如何评估最适合您的用例的 LLM 模型？ - Question 2: How to evaluate the RAG-based system?
问题 2：如何评估基于 RAG 的系统？ - Question 3: What are the different metrics that can be used to evaluate LLM
问题 3：有哪些不同的指标可用于评估 LLM - Question 4: Explain the Chain of verification问题 4：解释验证链
Hallucination control techniques幻觉控制技术
- Question 1: What are the different forms of hallucinations?
问题 1：幻觉有哪些不同形式？ - Question 2: How do you control hallucinations at different levels?
问题 2：你如何控制不同层次的幻觉？
Deployment of LLM部署 LLM
- Question 1: Why does quantization not decrease the accuracy of LLM?
问题 1：为什么量化不会降低 LLM 的准确性？
Agent-based system基于代理的系统
- Question 1: Explain the basic concepts of an agent and the types of strategies available to implement agents.
问题 1：解释代理的基本概念以及可用于实施代理的策略类型。 - Question 2: Why do we need agents and what are some common strategies to implement agents?
问题 2：为什么我们需要代理，实施代理的常见策略有哪些？ - Question 3: Explain ReAct prompting with a code example and its advantages
问题 3：通过代码示例解释 ReAct 提示及其优势 - Question 4: Explain Plan and Execute prompting strategy
问题 4：解释 Plan 和 Execute 提示策略 - Question 5: Explain OpenAI functions with code examples
问题 5：通过代码示例解释 OpenAI 函数 - Question 6: Explain the difference between OpenAI functions vs LangChain Agents.
问题 6：解释 OpenAI 函数与 LangChain 代理之间的区别。
Prompt Hacking提示黑客攻击
- Question 1: What is prompt hacking and why should we bother about it?
问题 1：什么是即时黑客攻击，我们为什么要为此烦恼？ - Question 2: What are the different types of prompt hacking?
问题 2：提示黑客攻击有哪些不同类型？ - Question 3: What are the different defense tactics from prompt hacking?
问题 3：与快速黑客攻击有哪些不同的防御策略？
Case study & scenario-based Question案例研究和基于情景的问题
- **Question 1: **How to optimize the cost of the overall LLM System?
问题 1：如何优化整个 LLM 系统的成本？
参考
https://github.com/NirDiamant/RAG_Techniques
https://medium.com/towards-artificial-intelligence/the-best-rag-stack-to-date-8dc035075e13
https://medium.com/@AMGAS14/list/natural-language-processing-0a856388a93a