Intro to Triton with Matrix Multiplication
Introduction to GPU programming with Triton and build the matrix multiplication along the way
Read More →Xiaoyou Wu
Computer Engineering @GaTech | Research in Efficient LLM & robot localization
Welcome to my research portfolio. I am a Computer Engineering student at Georgia Tech, passionately exploring efficient LLM inference, signal processing and high-speed PCB design
Research Interests
- Efficient NLP : LLM Quantization, Diffusion LLM, Speculative decoding
- Robot Localization: Particle Filters, SLAM
- Advanced PCB Design: High-Speed Layout, Signal Integrity, analog circuit design
- Signal Processing and Control theory: root locus, Autocorrection, Nyquist Plot
Currently Thinking
-
The relationship between transfer function and electric field?
-
How does Diffusion Blocks changes the inference and what factor determine the most efficient block size ?
-
Why Circuits have relatively larger LTI range than other field (language, material science, etc.)?
-
How to develop sharp intuition on the electrical components work ?
Featured Research Ideas
Quantization Aware Training Implementation Guide
Implementation details and best practices for Quantization Aware Training (QAT) with LoRA, including GPU memory optimization strategies
Read More →Quantization Strategy Report: Switchable and Cyclic Precision Training
Comprehensive report on quantization strategies including switchable precision and cyclic precision training applied to WikiText-103 dataset
Read More →