Tag
A user benchmarks Qwen 35B-A3B (a 35B MoE model) on a 12GB RTX 3060, finding that 12GB VRAM is a practical sweet spot for running the model with 32k context, achieving ~47 t/s generation.