2026-11-05 · America/Los_Angeles · 社区动态 · #1

Qwen3.8-27B has become unuseable recently

When it came out, I downloaded a Q4 quant of qwen3.8-27b and ran it using llama.cpp on my RTX 4090 and it was incredible. It could handle practically anything I threw at it, the most complex tasks and bugfixes. But lately it just keeps messing up and failing at the most basic of tasks or just straight up forgetting instructions. I'm telling you, they quantized it. They probably secretly nerfed it and made it worse so qwen4 looks better. And it got lazy too, now it doesn't want to finish the more complex tasks. I'm tired of companies doing this, I want my money back.

热度 46.9 / 100；排名与评分保留该期记录。