Blog | Feixiang Tao

Search the archive

Type to search titles, descriptions, and topics.

0 resultsCtrl / ⌘ K · Esc to close
local information: Shenzhen, CN
color palette:

Blog

Research, experiments & everyday observations.

3 entries
All
2026

关于 OPD 梯度的一个简单分析

8 min read
  • LLM
  • OPD
  • RL
  • Optimization
 2026

从一个 toy model 出发的 Transformer 机制解释性

19 min read
  • Transformer
  • Mechanistic Interpretability
  • Training Dynamics
  • TinyStories
  • Attention
 2026

低精度不应该用flops来进行计算强度的分析

9 min read
  • GPU
  • Roofline
  • FP16
  • Performance
  • CS336