fast neural machine translation on premise with Systran @pdhowler
fast neural machine translation on premise with Systran  @pdhowler
Uploaded October 2024 | Updated September 2026, 1 day ago
This shows NMT powered by Systran Translation Server installed in a Hyper-V VM on WIndows 11 Laptop. Even with just 8 CPU cores you can get very respectable performance. Ideally 16 cores or more,. 12 GB RAM as a minimum.

We've seen running it even on older PCs with DDR3 or DDR4 RAM at 2666 - 4000 MHz. The faster the RAM the better. If you have more RAM that also helps but it's not just the quantity of it, but also the speed of it.

In this example we're on a Lenovo laptop with 32 GB of RAM but only about 16 GB dedicated to the Virtual Machine. The memory is very fast at 6400 MHz DDR5.

This example had just 1 instance of each engine, one for English to French, another for English to Spanish. When both wre translating simultuneously, 16 cores get busy (8 per instance). In this simple demo however we only did the translations in a serial way, so only 8 cores are used, one engine at a time looping through 1000+ text files.

The text is from FBI interrogations during the Chris Watts murder investigation. It was produced via ASR from the available video recordings.
fast neural machine translation on premise with SystranDogwaffle does 3D   part 12b   Spritesheets and flying elephants in Spaceteaser - from chocolate coffee to Mars and beyondText Contour FillDogwaffle does 3D  - part 2  -  NASA ModelsNew image analysis tool (June 2026)MyMedia13b - sample animated Sky filter, with slight postNoise Menufrom black to blue: replacing the background colorHowler 2024 VideoTools - the TeaserGuildedTest - GO (  making shiny text )Dogwaffle does 3D- part 14 - OBJ files from Sloyd Generative AI - Automagic!
pdhowler |

fast neural machine translation on premise with Systran

SHARE TO X SHARE TO REDDIT SHARE TO FACEBOOK WALLPAPER