Jędrzej Maczan

Hi, it's Jędrzej

machine learning research

GitHub | Google Scholar | X | ORCID

publications

Full Paper

2026-07-31
Measuring and Reducing WebGPU Dispatch Overhead for LLM Inference
The Fourth UK AI Conference 2026
Nottingham, UK
OpenReview
Poster

2026-07-11
Memory-Efficient Activation Checkpointing with Sliding Window and Hirschberg's Algorithm for 0/1 Knapsack Solving in PyTorch
Fast Machine Learning for Science Conference 2026
University of California, San Diego, USA
Indico CERN · OpenReview
2026-02-09
Characterizing WebGPU Dispatch Overhead for LLM Inference Across Four GPU Vendors, Three Backends, and Three Browsers
arXiv
Open source

workshops / talks

Workshop Paper

2026-10-06
Dispatch Overhead, Not Kernel Quality: The Performance Bottleneck in Single-Stream WebGPU LLM Inference
COLM 2026 Workshop on Methods and Opportunities at Small Scale (MOSS)
San Francisco, USA
OpenReview
Poster

2026-09-14
"As a Language Model...": Chat Template Switches LLM Self-Referential Voice and Activation Steering Reproduces It
KONVENS 2026 First Workshop on Evaluating LLMs for Specialized Domains (Eval4SD)
University of Hamburg, Germany
OpenReview
Talk

2026-06-05
Online Softmax
Cohere Labs - Open Science ML Math Community

articles

2026-07-27
The cuBLAS transposition trick
Paged Out!, Issue #9, page 14
Open source
2026-05-29
From Boltzmann-Gibbs distribution to numerically stable online softmax on GPU with proofs
2026-02-19
Solving 0/1 Knapsack problem with sliding window and Hirschberg algorithm
Paged Out!, Issue #8, page 8
Open source
2025-11-21
Use 20x less peak RAM with dp_knapsack_sliding_hirschberg, a new activation memory budget solver for PyTorch
Open source
2025-10-01
Self-contained handwritten digit recognizer
Paged Out!, Issue #7, page 11
Open source
2025-03-01
A primer on Differentiable Architecture Search
Paged Out!, Issue #6, page 5
Open source
2024-12-01
GPT in PyTorch
Paged Out!, Issue #5, page 6
Open source
2024-06-01
Building automated machine learning with type inference
Paged Out!, Issue #4, page 4
Open source
[In progress]
tiny-vllm: Build your own high performance LLM inference engine in C++ and CUDA - a smaller version of vLLM
GitHub
Open source

about me

I like computers!

I want to be a full time research scientist and work on the intersection of math, AI and low-level systems
I worked as software engineer for 7 years and then as AI engineer for 2 years
My name is spelled as yenjay in English and 延杰 in Chinese
Email: jedrzej@maczan.pl
I'm doing Foundations of Probability and Statistics Specialization at University of Colorado Boulder, to get admitted to Master's program in AI
I got bachelor degree in computer science from Wrocław University of Science and Technology
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)

.
  .
...