Lossless abliteration of Qwen3.6-27B with NVFP4 hardware quantization for DGX Spark / Blackwell. BF16 (51 GB) + NVFP4 (26 GB) deployment guide, docker-compose, and QuickStart. - View it on GitHub