# WideNDepth - Show and tell > Source: > Published: 2026-09-20 22:21:24+00:00 Hello! I wanted to share a project I have been working on for a while, **WideNDepth (WND)** You can find the discussion at [widendepth-a-separation-of-knowledge-storage-from-iterative-reasoning](https://discuss.huggingface.co/t/widendepth-a-separation-of-knowledge-storage-from-iterative-reasoning/177809) Thanks to [@John6666](https://discuss.huggingface.co/u/john6666) for their feedback throughout the architecture development. WND is an experimental neural architecture I have been working on to separate reasoning from knowledge. You can review the architecture at **[WideNDepth](https://github.com/MB-Bidram/WideNDepth/)**. The basic idea is to have different parts of the model focus on different things, with the **Wide part** acting as a memory, and the **Depth part** acting as a reasoner. WiND is a minimal PyTorch framework for WideNDepth models. You can find the project here: [WiND](https://github.com/MB-Bidram/WiND) The library and the surrounding work are open source. If you find the project interesting, you’re welcome to look through it, experiment with it, change it, improve it, or build something new from it according to the project’s license. I cannot present WND and its small ecosystem as a finished solution, or claim that it is better than existing approaches in every situation. It’s an experimental project, and there is still room for more improvement. I would like to see if anyone finds the library and architecture interesting. This is mainly a **Show and Tell** to share the project I have spent time building and experimenting with, and hopefully give other people something interesting to explore. Please keep the original attribution and make it clear when something has been modified or is a separate project. The name and idea are not the same as [Google’s Wide & Deep](https://research.google/pubs/wide-deep-learning-for-recommender-systems) architecture.