NVIDIA Merlin is an open source library providing end-to-end GPU-accelerated recommender systems, from feature engineering and preprocessing to training deep learning models and running inference in production.
[ ] The >> operator, NVTabular workflow, Graph, fit/transform.
[ ] Training a ranking model: NVT + DLRM (or similar)
[ ] Training a retrieval model: NVT + TT (or similar)
[ ] Creating a basic System: Fetch Features -> NVT -> Ranking -> SampleSoftmax. (Since there's no retrieval model, input can be something like {user_id: u, item_ids: [a, b, c, d]}) @nv-alaiacano
The goal of this RMP is to revamp the "Getting Started" section of our Merlin documentation.
The introductory tutorials are meant to cover a limited number of things:
Introduce Merlin-specific syntax
Show how to perform common recsys tasks (train well-known model types) in Merlin
Introduce the concept of Systems with a basic example.
[ ] #1033
[ ] Installation guide: Docker, Conda, Pip, Launchpad, Google Colab @bschifferer
[ ] New tutorials with some "basics"
>>operator, NVTabular workflow, Graph, fit/transform.{user_id: u, item_ids: [a, b, c, d]}) @nv-alaiacano