Jonathon Wandy @RelentlessMon

Grad 2024 Calgary, Alberta Joined October 2023

Tweets

9
Followers

0
Following

30
Likes

6

Jonathon Wandy @RelentlessMon

a year ago

@Muennighoff @iclr_conf OLMoE is great. I'm hoping you guys do another open MoE soon. Also, Is there any chance that expert parallelism gets implemented for OLMoE? Or do you have any suggestions for how I might implement it?

0 0 0 17 0

View Details

Jonathon Wandy @RelentlessMon

a year ago

@zkwthu @JiachengYe15 I'm curious, what results do you think would be different if they did not use time t as an input into their architecture (your comprehensive work and RADD, I believe, has shown this parameter was unnecessary in discrete diffusion models)?

1 0 0 57 0

View Details

Jonathon Wandy @RelentlessMon

a year ago

@zkwthu @StanJames5038 Wow... I suppose there is a lot of opportunity then. From first glance, it looks like it's taking a a course to fine grain generation approach.

1 0 1 40 0

View Details

Jonathon Wandy @RelentlessMon

2 years ago

@aaron_lou @chenlin_meng @StefanoErmon Any plans on implementing FSDP into score entropy discrete diffusion? I want to train a non-pathetic model on my 3090s and would be glad to pay for you to implement it (I sent you a DM)

0 0 0 108 0

View Details

Jonathon Wandy @RelentlessMon

3 years ago

@NPCollapse @Lordcostanza Everyone can see it, except you. Clearly you must be the genius and everyone else is blind! You really think that your displays of contempt were productive in any way? Keep rationalizing buddy. Grow up.

0 0 1 72 0

View Details

Jonathon Wandy @RelentlessMon

3 years ago

@DylanoA4 This misses the mark by a mile. Humor has its time and place. And so does seriousness. The hero archetype is not a clown. And that's not an accident, bucko.

0 0 0 25 0

View Details

Jonathon Wandy @RelentlessMon

3 years ago

@DylanoA4 Me too. I'm not joking either. I think you're onto something. Elaborate if you may.

1 0 1 40 0

View Details

Jonathon Wandy @RelentlessMon

3 years ago

@DylanoA4 It is most definitely. Try again. For example: thefp.com/p/the-case-for…

0 0 0 16 0

View Details

Jonathon Wandy @RelentlessMon

3 years ago

@BrianJJi @Deepneuron No. It's much more banal than that. Speed is a competitive advantage. If you're aiming for a monopoly--you need every competitive advantage. "But why would Elon be aiming for a monopoly?" Read Zero to One. Elon recommends Zero to One.