Train a 4096x4096 Neural Network with NectarJS, something you can't with Node.js

6 points

6 years ago

Hi all,

NectarJS is becoming more mature every day : the memory is managed in real time without garbage collector and the performance are good.

To demonstrate what we can do, Wesley (from Discord) created some neural networks examples. The 4096 x 4096 nn example gives a JavaScript heap out of memory with NodeJS, but compiles and run fine with NectarJS : https://i.imgur.com/4BjDAgM.jpg

You can find the code here: https://github.com/NectarJS/nectarjs/blob/master/example/nn4096.js

Right now, NectarJS supports most of ES3, but we need help to improve NectarJS and make it better:

- Improve the std libs - Libuv integration for NodeJS support (started with fs.readFile as a POC and working fine) - Custom parser to self compile NectarJS - Supporting fully ES3, ES5, ES6 ...

NectarJS is fully open source, so don't hesitate to join us on Discord (https://discord.gg/cpe2UuN) or IRC (https://kiwiirc.com/client/irc.freenode.net/#nectarjs) and Github to (https://github.com/nectarjs/nectarjs) to help us to make it better.

Thx