LLM 25-Day Course

A comprehensive overview of key terms from AI, ML, DL differences to parameters, epochs, and loss functions

Key NLP terms including tokens, corpus, embeddings, and attention explained with examples

Understand BPE and WordPiece algorithms and Word2Vec concepts, with hands-on practice using tiktoken

Explaining the Transformer's encoder/decoder structure, Self-Attention, FFN, and normalization with diagrams

Implement Q/K/V, Scaled Dot-Product, Multi-Head Attention, and positional encoding from scratch with numpy

Understanding the essence of LLMs from scaling laws and pre-training to parameter-scale characteristics and emergent abilities

Model selection guide for the latest OpenAI lineup, Responses API usage, and cost management tips

Explore the Claude model lineup selection criteria, Constitutional AI concepts, and Anthropic Messages API usage

Understand the evolution of the Llama family and run the latest open models locally

A comprehensive overview of open-source LLMs including Qwen, DeepSeek, Phi, Yi, and Korean-language models SOLAR and EXAONE

Learn how to process images and text together using commercial and open-source multimodal model families

A comprehensive look at the Hugging Face ecosystem including Hub, Spaces, Transformers, Datasets, and PEFT

From installation to hands-on experience with pipeline() for text classification, sentiment analysis, translation, and summarization.

Understand the key parameters of model.generate() and compare generation strategies including temperature, top_p, and top_k through hands-on practice.

Deep dive into encode/decode, special tokens, padding/truncation strategies, and chat_template with apply_chat_template().

Learn Zero-shot, Few-shot, Chain-of-Thought techniques, structured output (JSON), and 10 practical prompt tips.

Build a complete RAG pipeline from document loading to Retrieval-Augmented Generation using LangChain and ChromaDB.

Understand the difference between Full Fine-Tuning and Parameter-Efficient FT, and learn decision criteria for fine-tuning vs prompting.

Learn the principles of LoRA (Low-Rank Adaptation), rank/alpha parameters, and QLoRA's 4-bit quantization approach through code.

Hands-on practice with LoraConfig setup, target module selection, trainable parameter verification, and model saving/loading using the PEFT library.

Learn Alpaca and ShareGPT dataset formats, data collection/cleaning/validation methods, and datasets library usage.

Execute actual training with trl library's SFTTrainer, and cover wandb monitoring and checkpoint management.

Compare precision levels from FP32 to INT4, explore GGUF/AWQ/GPTQ formats, and learn quantization methods with bitsandbytes and llama.cpp.

Serve LLMs locally with Ollama, vLLM, Text Generation WebUI, and llama.cpp, and compare their performance.

Integrate everything learned throughout the course: from model selection to fine-tuning, RAG, serving, and Gradio UI.