Blog

Last time we have shown how to run (or more like crawl) LLM inference on common CPUs. It can be rather sluggish though. And that is why we dug deeper and tried to use a commodity hardware - a gaming laptop - to speed things up! It is actually very interesting how even a gaming GPU can increase performance of AI tasks.

One of our current projects has to use LLM for extracting structured data from rather unstructured and noisy input. Such endeavor typically requires specialized GPU for decent inference times and it would be wise to test the solution with cheaper setup before buying such expensive hardware. It turns out it is possible to test even large models on CPU.

A few weeks ago, we published implementation of darkmode for static website without any javascript part 1, and part 2. This week, we would like to show you, how this useful feature can be implemented in any React application. Good news is, that coding such feature in React is way easier and requires less code!

Gracefully handling non-existent pages in any web presentation is usually a desired course of action. A bare-bones default "404 Not Found" page tells the user enough, but - frankly - does not look that exciting. But is it possible to have a nice multilingual error page while keeping everything tidy in a Puppet configuration like we use?

As a part of our work on new Key Management System (KMS) for rclone, it was more than appropriate to look for prior art. There are well-tested cryptosystems providing file encryption for file systems built on top of block devices. However it has turned out that these systems offer little to no inspiration for an object storage systems encryption.