LLMs

LLMs (with questoins)

• Core Concepts
  • Understanding LLMs
    • Architecture
      • Transformers
        • What is the core mechanism of a Transformer?
        • Explain the role of multi-head attention.
      • Attention Mechanisms
        • How does self-attention work?
        • What are the limitations of the self-attention mechanism?
      • Positional Encoding
        • Why is positional encoding necessary in Transformers?
      • Model Variants (Encoder, Decoder, etc.)
        • What are the key differences between encoder and decoder models?
    • Training
      • Pretraining
        • What is pre-training, and why is it crucial for LLMs?
      • Fine-Tuning
        • Explain the concept of fine-tuning LLMs.
      • Supervised Methods
        • When would you choose supervised fine-tuning?
      • Unsupervised Methods
        • What are unsupervised methods for LLM training?
    • Tokenization
      • Vocabulary
        • What is a token in the context of LLMs?
        • How does vocabulary size affect LLM performance?
      • Token Types
        • What are common types of tokens in language models?
      • Encoding Schemes
        • What is the difference between BPE and WordPiece tokenization?
    • Inference
      • Decoding Strategies (e.g., Greedy, Beam)
        • Compare and contrast greedy and beam search.
        • When would you use each decoding method?
      • Temperature and Sampling
        • What does the 'temperature' parameter control in LLMs?
      • Stopping Criteria
        • How do you define stopping criteria in LLMs?
    • Limitations
      • Hallucination
        • What is hallucination in LLMs, and how can you detect it?
      • Bias
        • How can LLMs exhibit bias, and what can we do about it?
      • Context Window
        • What is the context window of an LLM, and why is it important?
  • Information Retrieval & Knowledge Representation
    • Vector Embeddings
      • Semantic Representation
        • What are vector embeddings, and why are they important?
        • How do they represent meaning?
      • Embedding Models
        • What is an embedding model?
        • How do you choose the right embedding model?
      • Distance Metrics
        • What distance metrics are used to compare embeddings?
    • Vector Databases
      • Indexing Techniques
        • Explain different indexing techniques in vector databases.
        • When would you choose each one?
      • Similarity Search
        • How does a vector database perform similarity searches?
      • Filtering
        • What are challenges associated with filtering in a vector DB?
    • Chunking
      • Data Segmentation
        • Why is chunking data necessary in RAG?
      • Chunking Strategies
        • What are different chunking strategies?
      • Size Optimization
        • How do you determine the ideal chunk size?
    • RAG
      • Retrieval Techniques * How does the retrieval component work in RAG?
      • Augmentation Techniques * What is the augmentation part of RAG?
      • Hybrid Approaches * What are hybrid approaches in RAG and how do they improve results?
  • Prompt Engineering & Control
    • Prompt Design
      • Basic Structure
        • What are the basic components of a prompt?
      • Prompting Techniques (Few-Shot, CoT)
        • Explain few-shot prompting with examples.
        • What is Chain of Thought prompting?
      • Role-Playing
        • How can you use role-playing in prompting?
    • Prompt Optimization
      • Improving Reasoning * How do you improve reasoning ability through prompt engineering?
      • Handling Ambiguity * What prompt engineering strategies help with ambiguity?
      • Controlling Bias * How do you control bias with prompt engineering?
    • Hallucination Control
      • Prompt-Based Strategies
        • How to use prompts to control LLM hallucination?
      • External Knowledge
        • How can external knowledge reduce hallucinations?
      • Verification
        • How can verification techniques help reduce hallucinations?
• Advanced Techniques
  • Fine-Tuning & Adaptation
    • Supervised Fine-Tuning (SFT)
      • Dataset Creation
        • How do you create effective datasets for fine-tuning Q&A tasks?
      • Hyperparameter Tuning
        • How do you set hyperparameters for fine tuning?
      • Catastrophic Forgetting
        • What is catastrophic forgetting?
    • Preference Alignment
      • RLHF
        • What is RLHF and how is it used?
      • DPO
        • Explain how DPO works?
      • Reward Hacking
        • What is the reward hacking issue in RLHF?
    • Parameter-Efficient Fine-Tuning (PEFT)
      • Adapter Layers
        • What are adapter layers?
      • LoRA
        • Explain LoRA and how it works?
      • Prefix Tuning
        • Explain prefix tuning method
  • Advanced Search
    • Re-Ranking
      • Re-ranking Models * Why is re-ranking needed?
      • Fine-Tuning * How do you fine-tune re-ranking models?
    • Information Retrieval Metrics
      • Relevance Metrics * What are metrics used to evaluate relevance in information retrieval?
      • Accuracy Metrics * Which accuracy metrics should I use for information retrieval?
      • Ranking Metrics * How do you assess the quality of a ranking system?
    • Hybrid Search
      • Combining Methods
        • How does hybrid search work?
      • Homogenization
        • How do you homogenize results from multiple search methods?
  • Agent Systems
    • Agent Frameworks
      • Planning * What are basic concepts of agent planning?
      • Execution * What are basic concepts of agent execution?
      • Memory Management * What is agent memory and why is it important?
    • Tools & APIs
      • Function Calling * How do agents utilize function calling?
      • External Tools Integration * Why do we need to connect agents to external tools?
• Deployment & Evaluation
  • Deployment Strategies
    • Inference Optimization
      • Quantization * Why does quantization not decrease accuracy?
      • Model Pruning
        • How can we reduce model size using pruning?
      • Parallel Processing
        • How does parallel processing improve inference time?
    • Infrastructure Considerations
      • Cost Management * How do you optimize the cost of an LLM system?
      • Scalability * How do you design for scalable deployment?
      • Latency * How can you reduce latency of your LLM application?
  • Evaluation & Metrics
    • LLM Evaluation
      • Benchmarking * How do you benchmark your LLM for your use-case?
      • Human Evaluation * Why is human evaluation needed when evaluating LLMs?
      • Automatic Metrics * What automatic metrics can we use to evaluate LLMs?
    • RAG Evaluation
      • Retrieval Accuracy * How do you evaluate the retrieval component of your RAG?
      • Generation Quality * How do you evaluate the generated response of your RAG system?
• Security & Misc
  • Prompt Hacking
    • Types of Attacks
      • What are different types of prompt hacking?
    • Defense Mechanisms
      • What are some defense mechanisms against prompt hacking?
  • Miscellaneous Topics
    • Cost Optimization
      • How do you reduce the overall cost of your LLM system?
    • Mixture of Experts (MoE)
      • What is a Mixture of Experts model?
    • Hardware Considerations
      • How does hardware choices affect training and inference?
    • Low-Precision Training
      • How can we use low-precision training?