Qwen3.5 27B running at ~65tps with DFlash speculation on 2x 3090

r/LocalLLaMA
Generative AI

Submitted by /u/Kryesh [link] [comments]