Some news:
The original benchmark we ran on my m4 pro mac mini resulted in this video, which we thought was impressive:
~12.7 million queries per second is insane yes?
A few optimizations later, encoders now manages to hit ~30.6 million queries per second!
🏎️ 💨
If you think nanosecond-scale text embeddings are cool, wait until you see what’s possible with text embeddings in a hot path database transaction context
As a final note, you can read about encoders and play with a wasm embedding demo or two on our website:
https://t.co/EP3GeQa6F8
Shoutout to several people who inspired us and helped get this over the finish line in some way or another:
@neeks_at_nite + @jphorism + @tomaarsen + @minishlab + @tulkenss
If you’re resonating with Patrick’s description of liquid software, you might find our recent posts on the subject of liquid software, a term we use often, very interesting:
https://t.co/ZpR5btciHu
good piece from Flower on something increasingly important: agent coordination is becoming an infrastructure problem, not just an orchestration problem.
lots of interesting territory opens up from there, especially once coordination starts crossing data and trust boundaries.