maximal update parametrization (µP)
-
Updated
Jul 17, 2024 - Jupyter Notebook
maximal update parametrization (µP)
[NeurIPS 2025] Official implementation for our paper "Scaling Diffusion Transformers Efficiently via μP".
A port of muP to JAX/Haiku
Sheet music rendering plugin for Wordpress (UNMAINTAINED)
Maximal Update Parameterization in Tensorflow
Maximal-update parametrization coordinate checks and learning-rate transfer across model widths
Proteus Schematic Design and Assembly Code for a simple microwave
A fan control system, powered by assembly
Dockerfile for mup environment on arvm7
OTPs have long been identified as perfect encryption, however, they have limitations that have made them impractical. MUPs (Multiple Use Pads) are reusable OTPs that are practical and ideal. This is a basic example to demonstrate that the removal of language and plaintext patterns.
nanoWM: a pose-conditioned causal Diffusion Transformer world model at nano scale, built to measure which training tricks (REPA, Muon, muP, diffusion forcing) pay off in 20 ticketed GPU-hours on Kaggle T4s. The deliverable is the measurement, not the model.
From-scratch GPT and BPE tokenizer in raw PyTorch, plus µP and mixture-of-experts scaling experiments on cloud GPUs. BPE beats character tokens by 31% in bits/char; µP keeps the best learning rate fixed across widths; MoE beats dense per FLOP but not per GPU-second.
To associate your repository with the mup topic, visit your repo's landing page and select "manage topics."