← sources

Qwen 3.8 next flash tuning

source: reddit (reddit:LocalLLaMA)published 2026-09-20 17:36ingested 2026-09-21 02:38

retained text

Hello people 

First of all, thanks Qwen for this model, it really rocks!!

Anyway, I have problem tuning it and make it stable. I have all the experts except the gram in the Vram, and still the performance is not good. And the most annoying part is that the performance is qliff dropping as the context grow. if only it was stable as qwen 3.8...

Ok if anyone has some llama flag that can spare to help im all eyes.

See yaa

document fetched 2026-09-21 00:36 · canonical https://reddit.com/r/LocalLLaMA/comments/1wlncy6/qwen_38_next_flash_tuning