Seungjun Lee — ML Researcher

Portfolio Writing

Writing

NLP Bascis

RL Based Post-Training Methods(Policy Gradient, PPO, RLOO, and GRPO)
How Much VRAM Do You Need When Training/Inferencing a Model?
4 Types of Normalization
MQA, GQA, MLA
KV Cache
What’s Inductive Bais in LLMs/CNNs
Core Probability Theories
How Positional Encoding Evolved in Transformers: From Sinusoidal Encoding to RoPE
Flash Attention
Transformer
Model Quantization

Paper Summaries

DeepSeek-R1
Llama 2
DeepSeek Code
Switch Transformer
Kimi 1.5
DAPO
Qwen 3
s1
Understanding R1-Zero-Like Training: A Critical Perspective

Kaggle Playground Series

Kaggle Playground Series — General Strategy Playbook

Select a post from the sidebar, or open RL Based Post-Training Methods(Policy Gradient, PPO, RLOO, and GRPO).