Unsloth claims 2.5x faster NVFP4 quants for Qwen3.6

Unsloth's NVFP4 quantizations for Qwen3.6 27B are 2.5x faster than NVIDIA's implementation, while 35B-A3B variants achieve 1.56x to 1.79x speedups. The optimizations use W4A4 tensor cores with no accuracy degradation.