Qwen3.5 27B running at ~65tps with DFlash speculation on 2x 3090
r/LocalLLaMA
•
Generative AI
Submitted by /u/Kryesh [link] [comments]