Jędrzej Maczan

Hi, it's Jędrzej

ML systems researcher and engineer

writing

The cuBLAS transposition trick 2026-06-01
Paged Out!, Issue #9 (pending) · Open source

math linear-algebra cuda llm
torch-webgpu: PyTorch compiler and WebGPU runtime 2026-04-30
arXiv (pending) · Open source

c++ webgpu llm research
tiny-vllm: Build your own high performance LLM inference engine in C++ and CUDA - a smaller version of vLLM 2026-04-30
GitHub · Open source

c++ cuda hpc llm
Self-Reference in Base vs Instruction-Tuned Large Language Models 2026-04-14
arXiv (pending) · Open source

llm research
Characterizing WebGPU Dispatch Overhead for LLM Inference Across Four GPU Vendors, Three Backends, and Three Browsers 2026-04-05
arXiv · Open source

llm research webgpu performance pytorch torch-webgpu
Solving 0/1 Knapsack problem with sliding window and Hirschberg algorithm 2026-02-19
Paged Out!, Issue #8, page 8 · Open source

algorithms optimization math pytorch
Use 20x less peak RAM with dp_knapsack_sliding_hirschberg, a new activation memory budget solver for PyTorch 2025-11-21
Blog post · Open source

pytorch algorithms compilers optimization
Self-contained handwritten digit recognizer 2025-10-01
Paged Out!, Issue #7, page 11 · Open source

c ml neural-networks from-scratch
A primer on Differentiable Architecture Search 2025-03-01
Paged Out!, Issue #6, page 5 · Open source

ml automl neural-networks pytorch
GPT in PyTorch 2024-12-01
Paged Out!, Issue #5, page 6 · Open source

pytorch llm transformers
Building automated machine learning with type inference 2024-06-01
Paged Out!, Issue #4, page 4 · Open source

automl ml python

about me

I like computers!

Nowadays I focus on low-level deep learning - ML compilers and systems
My fav tech: C++, CUDA, Python, PyTorch, Linux, ML compilers, LLMs, Rust, WebGPU and TypeScript
I'm a regular author for Paged Out! magazine, which is a cool nerdy zine for things like hacking, programming and everything computer-related
I like math and I try to find time every week to learn something new. I'm currently slowly working through the "How to prove it" by Daniel Velleman (basic math undergrad book)
I want to be a full time research scientist and do research on the intersection of math, AI and low-level systems
My name is spelled as yenjay in English and 延杰 in Chinese
Night job:
- tiny-vllm - High performance LLM inference engine in C++ and CUDA, a younger sibling of vLLM, with continuous batching and paged attention
- torch-webgpu - PyTorch compiler and WebGPU runtime
- Research - LLMs, ML systems
Day job: Applying LLMs to things c:
GitHub: @jmaczan
Email: jedrzej@maczan.pl
I got bachelor degree in computer science from Wrocław University of Science and Technology. I'd like to get back to university one day and do master's/PhD in math or computer science! :D
Art of Anita: anitamaczan.pl, instagram: @anita_maczan
Other links (not very active):
- Mastodon: mathstodon.xyz/@jmaczan
- LinkedIn: click
- X/Twitter: @jedmaczan
- HackTheBox: @7563687575 (inactive, had fun some time ago)