Catching up on NLP and LLM (Part I)
As a Data Scientist, I actually have never had the chance to properly explore the newest progress in Natural Language Processing. With the summer and the brand new boom of Large Language Models for the reason that starting of the 12 months, I made a decision it was time to dive deep into the sector and embark on some mini-projects. In any case, there’s never a greater solution to learn than by practicing.
As my journey began, I noticed it was complicated to seek out content that takes the reader by the hand and goes, one step at a time, towards a deep comprehension of recent NLP models with concrete projects. That is how I made a decision to begin this latest series of articles.
Constructing a Comment Toxicity Ranker Using HuggingFace’s Transformer Models
In this primary article, we’re going to take a deep dive into constructing a comment toxicity ranker. This project is inspired by the “Jigsaw Rate Severity of Toxic Comments” competition which took place on Kaggle last 12 months.
The target of the competition was to construct a model with the capability to find out which comment (out of two comments given as input) is essentially the most toxic.
To accomplish that, the model will attribute to each comment passed as input a rating, which determines its relative toxicity.
What this text will cover
In this text, we’re going to train our first NLP Classifier using Pytorch and Hugging Face transformers. I is not going to go into the main points of how works transformers, but more into practical details and implementations and initiate some concepts that will probably be useful for the subsequent articles of the series.
Specifically, we are going to see:
- The way to download a model from Hugging Face Hub
- The way to customize and use an Encoder
- Construct and train a Pytorch ranker from one among the Hugging Face models
This text is directly addressed to data scientists that may wish to step their game in NLP from a practical perspective. I is not going to do much…