this shatters my mind.

[milorddev]

Holy crap dude. This is absolutely insane and incredible. Blown away. amazing job!
Going to try this on my 3080, see if Qwen3.6-35B-A3B works!

[Randozart]

Hah, I'm glad to hear you like it! Please let me know about the results once you've tried it.

[Randozart]

@milorddev So, you might like to know I have been integrating a custom coding runtime into this called Officina to run it locally on my PC. I am happy to say it runs Qwen3.8 27b Q3_K_S with just an 82k context window, but manages to stretch that well over 1 million tokens, while keeping context at 30k. And it's staying smart enough to do low level systems programming. Haven't tested it elsewhere yet.