Qwen 3.6 quants achieve 2.5x speedup, near-100K context on 32GB

Unsloth's NVFP4 quants for Qwen 3.6 27B deliver 2.5x speedup over NVIDIA's NVFP4 with no accuracy loss. Users report achieving near 100K context at Q8 on 32GB VRAM. vLLM benchmarks show NVFP4 fastest but with some stability issues.