FRAMEWIREIndonesiaUpdated Sep 14Live wire
0:00 / 0:00

Another test of my 2bit Qwen 3.8 27B.

A good example of the difference between standard 2-bit and our Opti compression on a task that is multi-faceted and requires different tooling. Standard llama.cpp IQ2_M (9.3 GiB): The “turtle” is basically a sparse point cloud over a grid,

kalebSep 14
0:00 / 0:00

This local AI can browse faster than you can!

This video is not sped up at all. It was made with BetterWright and Qwen 3.8 27b NVFP4. It did the task perfectly way faster than I ever could. 1-command adaptive local AI installation to BetterWright is now available in 2.8.0. It

SSHSep 141
0:00 / 0:00

I just made this game with Qwen 3.8 27B in less than 10 minutes on two R9700 AI Pros and it even made…

I just made this game with Qwen 3.8 27B in less than 10 minutes on two R9700 AI Pros and it even made the pixels of the character and the enemies just using math. AMD+vLLM+Radiance+Qwen 3.8 27B 🔥

CarloSep 1430
0:00 / 0:00

Yeah, it literally feels like AMD R9700 AI Pros are FLYING!

🔥 Using Radiance Quant: Qwen 3.8 27B MXFP4 by Just1

CarloSep 1482