Evidence receipt / belief
Published · transcript-backedVipul Ved Prakash: belief
8 Feb 2024 Latent Space Cloud Intelligence at the speed of 5000 tok/s - with Ce Zhang and Vipul Ved Prakash of Together AI
“I think on the inference stack, there are open source inference stacks which are pretty good and definitely today, it gives us a competitive advantage to have the best one.”
Source trail
Everything needed to verify it.
- Speaker
- Vipul Ved Prakash
- Attribution
- Verified speaker
- Claim type
- belief
- Recorded
- 8 Feb 2024
- Publisher
- Latent Space
Transcript context
…How much of these 50 tricks are you giving to yourself and how many are you gonna open? Because we have three now, obviously Flash Attention 2 is open source. He mentioned he'd love to come work together because of how much you care about open source. Yeah, how do you weigh that as a CEO and CTO? A lot of it is open, right? Flash Attention, Flash Decoding, et cetera, and we publish something that's very generally universally useful. It's going to produce better open source AI. We tend to publish as open source. I think on the inference stack, there are open source inference stacks which are pretty good and definitely today, it gives us a competitive advantage to have the best one. So we are not sort of rushing out to release everything about it. It's not overall that additive to open source out there and it is particularly useful as a business for us to provide best price performance. Yeah, we make these decisions. We have discussions. Anything that we keep closed, we generally talk about it quite a bit and decide like this is the piece that is closed for today and it may not be the case six months from now. It may not matter as much. Yeah, so I think being open is kind of very important, right? So I think the whole company actually built on this idea that there's going to be ecosystem built on our open models, right? And that's also how we are really lucky to attract this top group of talents to actually join us because of the dream and the mission that we have on our side to really facilitate the open ecosystem, right? So I think in general, it's like I think all the ideas should be open. So that's why we publish papers, right? We actually talk about ideas, right? So I don't think it makes any sense to keep idea like close, right? So there are some software artifact that are kind of really deeply embedded into our kind of own kind of like stack. It kind of only useful when you're trying to build a disaggregated cloud, right? Maybe at some point that we're going to be open as people said, right? But at this moment, right? So we are kind of busy actually building it, right? So that's probably kind of getting to the picture about when that piece is going to be open, right? But I think on the research side, the ideas and for our people to publish things, I think that's really, really important, right? So I think that's how we get talent. That's how I think we as a company going to move the field forward.…
Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.