2026-09-26 · America/Los_Angeles · 社区动态 · #6

Qwen 3.8 flash next is based on Qwen 4 architecture, if the announced Qwen 4 27b is also the same architecture with n-grams does it mean I can actually have faster inference on a single 3090 without tweaking much?

I wish Qwen also released dataset and method to fully train a model ourselves but it is what it is. However, I come here with my stupid question because someone can answer it better. And will the model still be an over thinker of faster inference will make up for that.

热度 47.5 / 100；排名与评分保留该期记录。