Projects

List of my projects from the most recent to the oldest.

Luth

Luth

Luth-0.6B-Instruct and Luth-1.7B-Instruct, two fine-tuned models trained on French data. They achieve SoTA results for their size on many French benchmarks, while still retaining competitive performance in English.

NLP MachineLearning Pytorch Python
LeCarnet

LeCarnet

LeCarnet is a synthetic dataset of 2M+ original French children’s stories using simple vocabulary. Trained and released LeCarnet-3M/8M/21M, a series of SLMs capable of generating fluent French stories despite their size.

MachineLearning Pytorch Python
FlashAttention-2

FlashAttention-2

Built FlashAttention-2 in Triton, including forward and backward pass for self-attention. Supports sliding window, causal and global attention.

Triton MachineLearning Pytorch Python
Mini-PaliGemma 2

Mini-PaliGemma 2

Minimalist implementation of PaliGemma 2 and PaliGemma VLM from scratch. Weights can be loaded for inference, and detections are displayed on input images.

MachineLearning NLP Pytorch Python
cWGAN-GP

Conditional WGAN-GP

Conditional WGAN-GP for graph generation based on descriptions for a Kaggle competition.

MachineLearning Pytorch Python
Word2Vec

Word2Vec

Word2vec and a convolutional classifier using its embeddings to predict the sentiment of movie reviews. Implemented from scratch using Pytorch.

MachineLearning NLP Pytorch Python
Projects