Each pod has a limit to how much data it can store. We want to give an example of how many vector embeddings you can store in a single pod. Since the data size is affected by number of dimensions, we have to choose a sample dimensionality for the example. We chose 768 since that was the most common last year (SBERT), although it might make sense to change it to 1536 (OpenAI).
I think it’s confusing because as an end user I don’t yet know what a pod is or why they’re needed. I’m guessing this is related to space partitioning?
It would probably be helpful to link to some docs on pods (are pods a leaky abstraction?) and first list the max size in number of floats or bytes and then say how many embeddings that would be with different models.
> Each s1 pod has enough capacity for 5M vectors with 768 dimensions.
I think that’s what I was remembering. I guess I just don’t understand what that means.