BentoML

Hi <@U04SNBBL3GV> - were you using `.bento` format for exporting or `.tar.gz`? cc <@U02HFFXSZ9D> any other suggestions?

This is unfortunately expected until we redo export with a more sensible format. Can you try using just `.tar` export instead?

<@UKB4CLKP1> thanks for the quick response. Yeah we are using the
```.bentomodel```
format.
We switched to the `.tar` format and have seen great improvements so far.
(cut down our export pipeline for sd2-1 from 8hs to 1h) :smile:

If you allow me a quick follow-up question: My current understanding is that it's expected to build the Container Images with the models stored inside them. The images can get quite large (wich isn't an issue for us yet). But Im just wondering if you're looking into alternative options like a remote model store?

user can already implement something like that in their code with external storage, and we’ve been thinking about the interface for users to do that more easily with the BentoML model store api, . however this will increase the startup time and makes scaling up a lot slower

I’m bentocloud, where we have more control, we are building this for all deployments by default to speed up things