NLP Products on a Raspberry Pi

📝 Blogs

A blog with a focus on Deep Learning, JVM and Performance.
Only last 4 blogs. All blogs↗

Seam Carving (Presentation & Workshop)
This is from a presentation I did last week (12th of May 2021). Notebook just under the slides! Please note that this requires the Kotlin kernel to run as it’s Kotlin and not Python.
More↗

Knowledge Distillation (Presentation)
Just a week ago I did a presentation on Knowledge Distillation and how it can help improve efficiency of models.
More↗

TIL: GitHub Pages + Progressive Web App (PWA) = ❤️
What Progressive Web Apps (PWAs) are web applications that are deployed on the Web and available on essentially every platform directly (Android, iOS, Linux, Windows, … you name it). They’re created the same way regular Web Apps are, with minor modifications like a new separate .js-file that runs a serviceWorker. This serviceWorker is the core of the PWA and handle things like caching & installation.
More↗

Building a Swedish Named Entity Recognition (NER) model (Flair/Huggingface)
Not interested in reading the whole article and just wanna play around with the model(s)? Head over to londogard.com/ner. P.S. The Flair model is available for simple installation through huggingface.co’s model hub
More↗

🤖 Projects

Presenting a few of our projects, find the rest here↗

SweNER

A State-of-the-Art Named Entity Recognizer for Swedish
Developed to strike a balance between performance, efficiency & usability there exists two models of different architectures.
The models are quantized (1/4th size, 4x faster) and run on a single Raspberry Pi along with our other projects!
Demo↗

ColorKidz

Convert your images into colouring pages
Our only Progressive Web App (PWA) this far. ColorKidz turns your photo into a colouring page using an algorithmic approach.
Demo↗

Summarizer
A State-of-the-Art language-independent extractive summarizer.
Built a very powerful yet simple summarizer that uses pure statistics, no large neural networks or anything. It does extractive summarization and it's only the TfIdf-variant is language independent.
Demo↗