往下拉回到首頁
天啊!這個 27B 的 Qwen 模型在 CloudeCode 上竟然跑得動,而且還滿順的?

天啊!這個 27B 的 Qwen 模型在 CloudeCode 上竟然跑得動,而且還滿順的?

Tried Qwen3.6-27B-UD-Q6_K_XL.gguf with CloudeCode, well I can't believe but it is usable

So I tried to run Qwen3-27B-UD-Q6_K_XL.gguf with 200K context on my RTX 5090 using llama.cpp. I'm getting around 50 tok/s, which is fine I guess, I don't really know this stuff so it might be improvable. But what I want to say is, I haven't tried local models for coding for quite a long time, and hell, I can't believe we're at the point where it's actually usable? Of course not the same first class experience as Opus 4.7, but damn, we are getting closer and closer. https://preview.redd.it/3pbvu