Joris Baan

transformer-from-scratch. Well documented, unit tested, type checked and formatted implementation of a vanilla transformer - for educational purposes.

DPAC-DialogueGAN. This repo implements GAN-based models for Dialogue Generation (DP-GAN, SeqGAN, and our own proposed DPAC-GAN)

ai-app-from-scratch. Designing, building, deploying, and scaling an AI-powered chat application from scratch. Full control over the language model, infrastructure, data and cost. Codebase and 2-part blog series.

calibration-on-disagreement-data. Code accompanying the EMNLP 2022 paper "Stop Measuring Calibration When Humans Disagree" in which we show problems with popular calibration metrics like ECE in settings where more than one answer is acceptable, and argue for several metrics that take into account the full human judgement distribution.

Natural-Gradient-RL. Project for the course Reinforcement Learning about natural gradients