Video of tensorfold running Qwen3.8-Flash-Next-MLX-oQ8-MTP high on M3 Ultra ~80 t/s! It's a great experience, less errors and better quality. Where possible I suggest to use the highest possible quants!
18
3
74
7,177
Any luck with 128GB or less?
1
1
151
Replying to @julianharris
DwarfStart + Q4 rocks!

Oct 3, 2026 · 10:43 AM UTC

1
135
Sort replies: Relevant Recent Liked
Replying to @ivanfioravanti
I need your recipe urgently :) What's the stack? I'll try it in my "build a full Miro clone" long-running benchmark. So far TensorFold has been regularly taking out my Mac requiring restarts so it's relegated to the horizon/ folder… github.com/boxabirds/awesome… -- I'd add it under combinations/ I have an M5 Max, AMD 395, and a 4090. Anything outside that I need some help :) DGX, 3090 etc…

GitHub - boxabirds/awesome-local-ai: Various configurations for optimal Local AI usage.

Various configurations for optimal Local AI usage. - boxabirds/awesome-local-ai

github.com
1
50