← Back to feed

Run 125B+ MoE on consumer GPUs

AI · Practice · 2026-11-12 10:53

开源项目Strata利用动态稀疏路由卸载技术,使Qwen3.8-Flash-Next等125B参数MoE模型能在单张12GB显存GPU上本地运行。

Source: stratallm.org


Auto-generated summary for reference only. RSS