Florian Hofmann

1. Ok ... that makes sense. Regarding fusion, is that implemented? Or are there plans for it? When there is no sharing between events this would probably allow other rewrite...

Ooook ... stupid me ... just implemented the foldTree for my vectors and now it's quite fast after the initial compilation phase. Guess that will probably solve my other problems...

Hi again! Roughly how much of a speedup can I expect from `reactive-banana-0.9`? Regarding fusion: I just did some experiments on how much events per second I can pump through...

> A significant one. Nice ... any change to get a pre-release version of that? ;) > For the moment, I'd say that we have to wait for 0.9 before...

I just realized: even with an optimizing `compile` function this won't fuse as this all happens at run-time ... long after rewrite rules are fired.

I kind of assume that OpenVINO uses CPU features / instructions that are only available once per core. Also, keep in mind that half of the threads are "just" hyperthreading,...

> release int8 version yet Please do (at your own leisure), I think it would be interesting to see how competitive current Gen Intel processors are in this field.

The original Laion5B dataset does have a "NSFW" score: > Additionally, we provide several nearest neighbor indices, an improved web interface for exploration & subset creation as well as detection...

This should get linked somewhere in the "common issuses" section in the wiki ... took me ages to find.

`nm cbits/hsstring.o` → http://pastebin.com/gZ4HwsCQ `nm cbits/hsstring.o | c++filt` → http://pastebin.com/a3Y0j6pf