What does that mean?

So, you know, today you can find dozens of open source models that do a variety of tasks.

And we have a lot of customers that come to us and say, what is the performance of model X or model Y on your chip, you know, in this use case? And a lot of times what we find is they start out with that because it's open source, they can get it, but it was really built for a different purpose.

It was built with big compute in mind.

then they're trying to jam it onto what we would traditionally call an embedded device maybe still a processor class device but still um you know they come in with these uh lofty expectations of hey we want to be able to do this amount of ai processing and something with this big envelope and my answer normally is well it's kind of the wrong approach you're taking a you're taking a swiss army knife into the problem uh when there's a variety of smaller models that can run very efficiently on the device.

And if we orchestrate them correctly, you get the same output.

So you can get a single model that can process a lot of different modalities.

Can you talk about that?