Skip to main content

Ctrl+K

AI-101

Transformers

Overview
Tokenizer
Embedding
Attention
Normalization
Feed-Forward Network (FFN) and SwiGLU
Positional Embeddings and Context Extension
Summary

Post-training

Post-training Overview
Supervised Fine-tuning
RLHF
- PPO From Scratch: From Expected Return to the Clipped Objective
- From Return to Advantage: Full Derivation to GAE
RLVR
- GRPO and Its Improvements: Vanilla GRPO -> DAPO, Dr. GRPO, and GSPO
Knowledge Distillation
TRL Examples
- Agentic RL with GRPO: Tools, Rewards, and Rollouts
- Multi-task Reward Training with GRPO

Evaluation

Evaluation Overview
OpenJudge Built-in Graders
LLM Evaluation Metrics
Inter-Annotator Agreement
OLMo 3 Model Flow Overview
OLMo 3 Evaluation Task Clusters

Responsible AI

Overview
OpenAI Safety Evaluations Hub
OpenAI Moderation API
GPT‑OSS‑Safeguard: Policy‑Conditioned Safety Classification
Anthropic - Protecting the Well-Being of Users

Miscellaneous

References
Changelog
Credits and License

Repository
Suggest edit
Open issue

.md

RLHF

RLHF#

Content coming soon.

previous

Supervised Fine-tuning

next

PPO From Scratch: From Expected Return to the Clipped Objective

By TeachBooks Team, built with TeachBooks, CC BY 4.0

Last updated on January 5, 2026.