deadmanoz on Nostr: Too bad it’s also really slow.. Chinese labs need to catch-up on inference ...

Too bad it’s also really slow..

Chinese labs need to catch-up on inference infrastructure for scale!!

quoting
nevent1q…zv86

It might be true..

I’ve been using Kimi occasionally for a few months, mainly only when I had no usage remaining on Claude/Codex, because it was much, much weaker

But K3 does actually seem to be a very capable model from my brief experimentation

Now if only I had the hardware to self host it 🥲