{"type":"link","version":"1.0","title":"该模型基于 26B MoE… | DiffusionGemma: 4x… · HotDaily 每日晚报","description":"该模型基于 26B MoE 架构（推理仅激活 3.8B），打破了传统 LLM 逐字生成的模式，可并行生成 256 个 token，在 H100 上速度超过…","url":"https://hotdaily.top/item/18faff4b-93b6-4d13-a056-e9e0b6b25475","provider_name":"HotDaily","provider_url":"https://hotdaily.top","thumbnail_url":"https://hotdaily.top/og.png","thumbnail_width":1200,"thumbnail_height":630}