[reddit r/LocalLLaMA] Qwen3.6-35B-A3B-APEX / 128K ctx on RTX 3060 12GB — 37 t/s gen with 72k ctx filled, PPL 3.25, offloading 17GB model