3 ms·
Qwen3.8-Next, thanks to its new architecture, is quite fast even if part of it is streaming from disk
by jokethrowaway 16d ago
Qwen3.8-Next, thanks to its new architecture, is quite fast even if part of it is streaming from disk
- kennywinker 16d agoTotally. Any MoE model can have experts swapped in and out from disk or system ram. I only framed it this way because the question was about the model fitting in vram.