Unsloth claims 2.5x faster Qwen3.6 NVFP4 quants

Unsloth's NVFP4 quants for Qwen3.6 27B achieve 2.5x speedup over NVIDIA's version with no accuracy loss. The 35B-A3B variant sees 1.56x to 1.79x speedup using W4A4 matmuls instead of W4A16.