# Generative AI (Gen AI)

![](/files/kS4aLS5h0GtX2BTntkE0)

* **Deep Understanding of Transformer Architecture** \
  Why attention > RNNs | Self-attention | Q, K, V intuition | Multi-head attention | Positional encoding | Causal masking | Cross attention
* **Large Language Model Training Concepts**                                                                                          Pretraining vs Fine-tuning | SFT | RLHF | DPO | Instruction tuning | LoRA / QLoRA | PEFT | Quantization (fp16, bf16, int8, fp8) |  Offloading | vLLM | Paged attention | KV cache
* **Tokenization.**                                                                                                                                                                      BPE | SentencePiece | Why tokenization matters | Context window impact | Token explosion problems
* **Inference Optimization**                                                                                                                                                                                Why vLLM > naive HF inference | Continuous batching | KV cache | Paged attention | Throughput vs latency tradeoff | GPU memory planning | Temperature | Top-k & Top-p | Beam search | Sampling strategies | Latency vs throughput tradeoff | GPU vs CPU inference

**GenAI System Design**

Multi-agent systems | Tool use (function calling) | ReAct pattern | Memory management | Long context handling | Guardrails

Retrieval layer | Tool layer | Safety layer | Monitoring | Prompt injection defense

**Notes :-** 

Speculative decoding | Mixture of Experts | Flash Attention | KV cache compression | Context length extrapolation (RoPE scaling) | Evaluation: BLEU is bad for LLMs | LLM as judge (pros & pitfalls)

* Why does RAG reduce hallucination?
* When would you fine-tune vs use RAG?
* How does LoRA work mathematically?
* What causes hallucination technically?
* How would you evaluate a GenAI system ?
* What is temperature doing mathematically?

How RAG & Agent LLM model works | RAG vs Agent | Shared memory between Agents |  Embedding models |  how to access a very big RAG

How do you deal with PDF's containing images in RAG |  What are the prompt engineering techniques |  Types of chucking | 


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://59r.gitbook.io/ml-university/generative-ai-gen-ai.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.