Qwen3.8-27B MTP Q4S head on Q6 quant is faster than q6, f16 slower = acceptance

(github.com) by xlayn | Nov 8, 2026 | 1 comments on HN