Speaker unverified: belief

Transcript context

…I know you mentioned some of the other inference providers. I think the other question that people always have is around benchmarks. So you get different performance on different platforms. How should people think about... People are like, hey, Lama 3.2 is X on MMLU. But maybe using speculative decoding, you go down a different path. Maybe some providers run a quantized model. How should people think about how much they should care about how you're actually running the model? What's the delta between all the magic that you do and what a raw model... Okay, so there are two big development cycles. One is experimentation, where they need fast iteration. They don't want to think about quality, and they just want to experiment with product experience and so on. So that's one. And then it looks good, and they want to post-product market with scaling. And the quality is really important. And latency and all the other things are becoming important. During the experimentation phase, it's just pick a good model. Don't worry about anything else. Make sure you even generate the right solution to your product. And that's the focus. And then post-product market fit, then that's kind of the three-dimensional optimization curve start to kick in across quality, latency, cost, where you should land. And to me, it's purely a product decision. To many products, if you choose a lower quality, but better speed and lower cost, but it doesn't make a difference to the product experience, then you should do it. So that's why I think inference is part of the validation. The validation doesn't stop at offline eval. The validation will go through A-B testing, through inference. And that's where we offer various different configurations for you to test which is the best setting. So this is the traditional product evaluation. So product evaluation should also include your new model versions and different model setup into the consideration. I want to specifically talk about what happens a few months ago with some of your major competitors. I mean, all of this is public. What is your take on what happens? And maybe you want to set the record straight on how Fireworks does quantization because I think a lot of people may have outdated perceptions or they didn't read the clarification post on your approach to quantization.…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.