ByteByteAI – Learn by Doing. Become an AI Engineer

LLM Overview and Foundations
Pre-Training

• Data collection (manual crawling, Common Crawl)
• Data cleaning (RefinedWeb, Dolma, FineWeb)
• Tokenization (e.g., BPE)
• Architecture (neural networks, Transformers, GPT family, DeepSeek, Qwen, Gemma)
• Text generation (greedy and beam search, top-k, top-p)

Post-Training

• SFT
• RL and RLHF (verifiable tasks, reward models, PPO, etc.)

Evaluation

• Traditional metrics
• Task-specific benchmarks
• Human evaluation and leaderboards

Chatbots’ Overall Design

Overview of Adaptation Techniques
Finetuning

• Parameter-efficient fine-tuning (PEFT)
• Adapters and LoRA

Prompt Engineering

• Few-shot and zero-shot prompting
• Chain-of-thought prompting
• Role-specific and user-context prompting

RAGs Overview
Retrieval

• Document parsing (rule-based, AI-based) and chunking strategies
• Indexing (keyword, full-text, knowledge-based, vector-based, embedding models)

Generation

• Search methods (exact and approximate nearest neighbor)
• Prompt engineering for RAGs

RAFT: Training technique for RAGs
Evaluation (context relevance, faithfulness, answer correctness)
RAGs’ Overall Design