{"slug": "widendepth-show-and-tell", "title": "WideNDepth - Show and tell", "summary": "An independent developer released WideNDepth (WND), an experimental open-source neural architecture that separates knowledge storage from iterative reasoning, with the \"Wide\" component acting as memory and the \"Depth\" component acting as reasoner. The project is accompanied by WiND, a minimal PyTorch framework for building WideNDepth models, and both are published on GitHub under an open-source license. The developer, MB-Bidram, stated the work is experimental and does not claim to outperform existing approaches in every situation, and noted the name and idea differ from Google's Wide & Deep architecture.", "body_md": "Hello!                                                                                                                                                                                                             I wanted to share a project I have been working on for a while, **WideNDepth (WND)**\n\nYou 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.\n\nWND 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/)**.\n\nThe 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.\n\nWiND is a minimal PyTorch framework for WideNDepth models.                                                                                     You can find the project here: [WiND](https://github.com/MB-Bidram/WiND)\n\nThe 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.\n\nI 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.\n\nI would like to see if anyone finds the library and architecture interesting.\n\nThis 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.\n\nPlease keep the original attribution and make it clear when something has been modified or is a separate project.\n\nThe name and idea are not the same as [Google’s Wide & Deep](https://research.google/pubs/wide-deep-learning-for-recommender-systems) architecture.", "url": "https://wpnews.pro/news/widendepth-show-and-tell", "canonical_source": "https://discuss.huggingface.co/t/widendepth-show-and-tell/180648#post_1", "published_at": "2026-09-20 22:21:24+00:00", "updated_at": "2026-09-20 22:24:33.241714+00:00", "lang": "en", "topics": ["neural-networks", "machine-learning", "ai-research", "developer-tools"], "entities": ["WideNDepth", "WiND", "PyTorch", "GitHub", "Hugging Face", "MB-Bidram", "Google", "Wide & Deep"], "alternates": {"html": "https://wpnews.pro/news/widendepth-show-and-tell", "markdown": "https://wpnews.pro/news/widendepth-show-and-tell.md", "text": "https://wpnews.pro/news/widendepth-show-and-tell.txt", "jsonld": "https://wpnews.pro/news/widendepth-show-and-tell.jsonld"}}