Benchmarking Qwen3.8 27B quantizations: 4-bit holds up, 1-bit collapses

How much GPU RAM do you actually need to run Qwen3.8 27B without sacrificing quality? The full BF16 model weighs 55 GB, putting it beyond most consumer hardware. Yet the 17 GB Q4_K_M matches the full

事件进展（1 篇报道）

1. 2026-09-08 22:49
   Hacker News Benchmarking Qwen3.8 27B quantizations: 4-bit holds up, 1-bit collapses

← 返回首页