RLHF and DPO: Aligning AI to Human Preferences
Comparison of RLHF and DPO alignment techniques, covering their pipelines, strengths, weaknesses, and where each is used in practice
Notes
Instruction Tuning
Overview of instruction tuning, how it works, dataset construction, and variants like RLHF, RLAIF, and DPO
Perplexity in Language Models
Explanation of perplexity as a language model evaluation metric, including the formula, intuition, caveats, and relationship to cross-entropy loss
Model Quantization
Overview of model quantization techniques, precision levels, and trade-offs for reducing neural network memory and improving inference speed
GCloud Quick Reference
Quick reference for common gcloud commands including authentication, project setup, and GKE cluster configuration
Kubernetes Port Forward
Quick reference for kubectl port-forward command syntax