A quick update on Dataset 3

Message boards : Science : A quick update on Dataset 3
Message board moderation

To post messages, you must log in.

AuthorMessage
pianoman [MLC@Home Admin]
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 30 Jun 20
Posts: 462
Credit: 21,406,548
RAC: 0
Message 450 - Posted: 3 Sep 2020, 4:16:40 UTC

We're running Dataset 3 WUs through testing, and I'm experiencing an issue where the networks aren't learning. This likely means there's a bug in my generation code, so I'm going back to the drawing board and running more tests before release, expect a longer delay for dataset 3 WUs.

In parallel, We're also working on Dataset 4, which will require a CNN network and train on MNIST-based datasets. I'm not sure which will be ready first.

Just wanted to give an update since the we wanted to have Dataset 3 WUs out two weeks ago, and they still aren't ready. There are things happening, its just not publically visible at the moment.
ID: 450 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
[VENETO] boboviz

Send message
Joined: 11 Jul 20
Posts: 33
Credit: 1,266,237
RAC: 0
Message 451 - Posted: 3 Sep 2020, 8:05:29 UTC - in response to Message 450.  

There are things happening, its just not publically visible at the moment.

No problem. Meantime we crunch Dataset 1&2!! :-)
ID: 451 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
bozz4science

Send message
Joined: 9 Jul 20
Posts: 142
Credit: 11,536,204
RAC: 3
Message 1280 - Posted: 26 Jul 2021, 8:52:00 UTC

I do see DS4 listed on the main page and that means that you are working hard on the launch of DS4 behind the scenes. Can you give an estimate for when you expect DS4 to launch on MLC? Thx
ID: 1280 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
pianoman [MLC@Home Admin]
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 30 Jun 20
Posts: 462
Credit: 21,406,548
RAC: 0
Message 1298 - Posted: 29 Jul 2021, 21:23:46 UTC

Currently working on getting the updated windows CPU client out the door to match the updated Linux client. Just last night after weeks of trying I finally coaxed pytorch to build statically on windows. Now I need to convince the client to link against it. As I mentione din another thread, this involves a lot of trial and error, building compiler link lines by hand until you get one that works, and then converting it back to cmake syntax for readability.

But DS4 itself it pretty much ready to go on the Linux side, I just need to push the test WUs. Maybe I'll get to some of those tonight.

For DS4, we'll be training both a dense network and some CNN networks with image classification datasets: 3 MNIST-like ones, and CIFAR-10 and CIFAR-100... both of which are about 160MB. These are industry standard benchmarks.

On the web page you see "Dense" and "LeNet5", but I also plan on having a few others as well, some substet of Imagenet, Alexnet, and/or resnet to round out the data set.

Feel free to follow along with development on discord.
ID: 1298 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
bozz4science

Send message
Joined: 9 Jul 20
Posts: 142
Credit: 11,536,204
RAC: 3
Message 1302 - Posted: 31 Jul 2021, 13:01:21 UTC

Thank you so much for taking the time to getting back to me. The whole development process sounds tiresome, especially that you have to build compiler link lines by hand! I will follow along the development process on Discord, but not really sure if I can be of much help. You could launch a DS4 assault on the Linux machines and release these DS4 WUs into the wild before you get the Windows client updated and ready to go. I am also way more excited for the launch of the DS4 experiment than I was at the launch of DS2/3. Implications for industry applications and research could be far more profound with these kinds of networks.

Do you have an estimate for the expected runtime of these networks? I guess that the input files are rather large, requiring more RAM and VRAM on the GPU, as well as the runtimes to be longer...

Appreciate your responsiveness as always and keep up the awesome work!
ID: 1302 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote

Message boards : Science : A quick update on Dataset 3

©2022 MLC@Home Team
A project of the Cognition, Robotics, and Learning (CORAL) Lab at the University of Maryland, Baltimore County (UMBC)