Hello, I’d like to introduce myself and a project I’ve been working on.
I’m a self-employed data scientist, with a particular interest in NLP. I’m interested in trying to make NLP models more factually reliable, so I’ve started a project to investigate some ideas I’ve had on the subject. It’s called QARAC, which stands for Question Answering, Reasoning and Consistency, which are the training objectives of my models.
I’m using RoBERTa as my base model, and was originally using Keras to implement it. However, my training system requires training 3 models together on 4 objectives, so I’ve ported it to PyTorch, which seems more suitable for that.