MNIST training: Showdown between JavaScript and WebAssembly

#1

Post by neptvn » Sun, Mar 11, 2018, 5:25 PM UTC

MNIST training: Showdown between JavaScript and WebAssembly
ai.danruta.co.uk

Re: MNIST training: Showdown between JavaScript and WebAssembly

#2

Post by catman99 » Sun, Mar 11, 2018, 5:34 PM UTC

If you care about performance use GPU shaders.

Re: MNIST training: Showdown between JavaScript and WebAssembly

#3

Post by dvddgld » Sun, Mar 11, 2018, 5:40 PM UTC

I compiled hello world in Rust into wasm the other day for the first time and it was incredibly satisfying for whatever reason. New technology is fun I guess.

Note: This page isn't rendering properly on iPhone

Re: MNIST training: Showdown between JavaScript and WebAssembly

#4

Post by nsthorat » Sun, Mar 11, 2018, 5:53 PM UTC

catman99 wrote: post #2

If you care about performance use GPU shaders.

https://github.com/pair-code/deeplearnjs

Re: MNIST training: Showdown between JavaScript and WebAssembly

#5

Post by nsthorat » Sun, Mar 11, 2018, 5:53 PM UTC

Come build a WASM backend for deeplearn.js :)

Re: MNIST training: Showdown between JavaScript and WebAssembly

#6

Post by darknoon » Sun, Mar 11, 2018, 5:56 PM UTC

It would be a more fair comparison to https://deeplearnjs.org/, since CPU training is not standard practice for neural nets these days.

Re: MNIST training: Showdown between JavaScript and WebAssembly

#7

Post by nsthorat » Sun, Mar 11, 2018, 5:58 PM UTC

darknoon wrote: post #6

It would be a more fair comparison to https://deeplearnjs.org/ , since CPU training is not standard practice for neural nets these days.

We've done some initial tests ourselves. WASM doesn't yet support SIMD so WebGL tends to be 5-10x faster. SIMD is actively being worked on by many smart people in Chromium / other browsers, so I would expect to see huge wins in the near term future. When that happens, deeplearn.js will have a WASM backend. WASM has a much better memory management story (destructors on the C++ side) so I'm super excited about its future.

Re: MNIST training: Showdown between JavaScript and WebAssembly

#8

Post by ekelsen » Sun, Mar 11, 2018, 6:06 PM UTC

catman99 wrote: post #2

If you care about performance use GPU shaders.

If you care about performance, don't use JS for training neural networks.

Re: MNIST training: Showdown between JavaScript and WebAssembly

#9

Post by akmittal » Sun, Mar 11, 2018, 6:34 PM UTC

I showed me 1 min as 31 mins

Re: MNIST training: Showdown between JavaScript and WebAssembly

#10

Post by lol768 » Sun, Mar 11, 2018, 6:36 PM UTC

My browser at least seems to struggle to download the file at https://ai.danruta.co.uk/webassembly/mnist.js - how big is this file in total?