ml-fairness-framework

GPT* - Training faster small transformers using ALiBi, Parallel Residual Connections and more!

Responsible AI Masterclass (June 2024 Run)

Imagine boarding the Titanic in 2021, and you have provided all your details as a passenger to the captain. There is are three people involved, the data scientist, captain and the passenger. Imagine the company who has built Titanic has created a machine ML model to predict the rate of survival of the passengers, in case of a disaster. The job of the data scientist is to make a model that is explainable to the passengers who are not technical, and that they get the answer about the reasons why they may not survive.

This notebook is ispired by the AIX360 HELOC Credit Approval Tutorial, which shows different explainability methods for a credit approval process. Here XGBoost is used for classification, achieving better accuracy than most of the models used in that notebook. Then, feature importance methods are shown, to be compared with the Data Scientist explanations methods provided in the above notebook. The first ones come directly with XGBoost and the other is based on SHAP.

Testing MLServer and Alibi

「Pytorch 기반 GPT 모델 및 모듈 라이브러리」에 대한 내용을 다루고 있습니다.

一个关注伦理的深度学习项目，旨在预测信用卡优惠的接受度，同时利用SHAP、Fairlearn和AIF360等工具来减轻基于收入的偏见。

Transformer Models for Humorous Text Generation. Fine-tuned on Russian jokes dataset with ALiBi, RoPE, GQA, and SwiGLU.Plus a custom Byte-level BPE tokenizer.