This repository walks through a few ideas of how to implement generative recommendations for friending recommendations and in doing so describes the complexity or what is uniquely difficult in frie…

We write sample code for two tower models for retrieval and add RLHF/RLAIF style alignment with a ranking model to make the retrieval more aligned with the ranking model on top

Using machine learning to decide when to show an ad

We demonstrate an extension of two tower models for recommendation that work jointly for search and recommendation

Recommending friends on a social network

nDCG and regret based tools to find value model weights

We show how to add a calibration arch to multi task multi-label models used in ranking of recommender systems

Describing the merit of constructing recommendation feeds that take into account risk

Multiple ways to model user preference in recommender systems

Covers a couple of approaches to training an early ranker with knowledge distillation from final ranker

We will show how to factorize and debias ranking to improve personalization and reduce popularity bias both on item side and to reduce the dominance of power users.

Adding REINFORCE based reward maximization to pointwise ranking

Pipelined early ranker in a recommender system

Trying to emulate Andrej Karpathy's language model video

Trying out how to use GPT-3 for a chatbot

Something that given a list of prohibited words, filters out text that contains them

Sample Go and Rust implementation of Vector Clocks

A sample round robin ranking implementation