insane. Competing with Qwen 3 (latest) while using ~6T tokens, and all in the open. They need to refine OLMoE recipe now, and they may well become a mainstream choice for research.
正在加载线程详情
正在从 X 获取原始推文,整理成清爽的阅读视图。
通常只需几秒钟,请稍候。
正在加载线程详情
正在从 X 获取原始推文,整理成清爽的阅读视图。
通常只需几秒钟,请稍候。
共 1 条推文 · 2025年11月20日 15:45
insane. Competing with Qwen 3 (latest) while using ~6T tokens, and all in the open. They need to refine OLMoE recipe now, and they may well become a mainstream choice for research.