Qwen 3.8 next flash tuning
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