Qwen3.8-27B uncensored Q6_K at 156K context on one RTX 5090, 140-190 tok/s with DFlash2

A user successfully ran Qwen3.8-27B RVN Heretic at Q6_K with 159744 context on a single RTX 5090, achieving 140-190 tok/s using DFlash2 speculative decoding. The setup included a vision projector on the GPU and a Sharp chat template, with measurements taken on 2026-09-16. Token embeddings remained on the host, accounting for the last ~1 GB of context.