Llama 4 - Scout: best quantization resource and comparison to Llama 3.3

The two primary resources I’ve seen to get for Scout (GGUF for us GPU poor), seems to be Unsloth and Bartowski… both of which seems to do something non-traditional compared to density models like Llama 70b 3.3. So which one is the best or am I missing one? At first blush Bartowski seems to perform better but then again my first attempt with Unsloth was a smaller quant… so I’m curious what others think. Then for llama 3.3 vs scout it seems comparable with maybe llama 3.3 having better performance and scout definitely far faster at the same performance.