Deep Dive Into LLMs Like ChatGPT
Master the architecture and training pipelines behind modern transformers to bridge the gap between high-level prompting and low-level model optimization for production-grade engineering and research. * **Master pre-training** and tokenization strategies for massive datasets * **Implement SFT and RLHF** to align model behavior * **Quantify model limits** through rigorous benchmarking and evaluation * **Architect RAG pipelines** to mitigate hallucinations and latency
5 sections ยท 20 lessons
Course outline
Training Pipeline
- Three Training Stages
- Internet Data Curation
- Tokenization Fundamentals
- Neural Network Training
- Base Models Explained
- Text Generation Process
- Training Pipeline
Post-Training
- Supervised Fine-Tuning
- Conversation Tokenization
Limitations
- Hallucinations
- Tool Use Integration
- Computational Limits
- Tokenization Blind Spots
Reinforcement Learning
- Reinforcement Learning Basics
- RL in Math & Code
- Chain of Thought
- RLHF for Subjective Tasks
- RLHF Limitations
Practical Insights
- Mental Model of AI
- LLM Ecosystem
- Future Capabilities