projects
research
neural_network_compression_patent
Methods to combine quantization, pruning and low-rank approximation for neural network compression. Developed at Thales under supervision of Benoit Porteboeuf.
tool
sqlgbm
SQL-based inference for gradient boosting models
api
invertornot.com
API to optimize image inversion for dark mode (processes 40K+ images/day)
oss
tinygrad_contributions
Added learning rate schedulers to the tinygrad framework
misc
ml_implementations
Implementation of different models (GAN, VAE, SRGAN...) and various exercises (Advent of Code, Cryptopals...)
engine
voxel_space_engine
Toy implementation of a voxel space engine using Zig