面壁智能关注 MiniCPM5-2B 本地延迟表现
@0xBakeer Appreciate the single Spark measurements. MiniCPM5-2B + drafter at 100.8 tok/s is the local latency profile we care about. Solid apples-to-apples on o
AI 导读
@0xBakeer 感谢提供单台 Spark 的测量数据。MiniCPM5-2B + drafter 达到 100.8 tok/s，正是我们关注的本地延迟表现。单机上的对比很扎实。
正文
当前显示原文，中文全文译文尚未就绪。
@0xBakeer Appreciate the single Spark measurements. MiniCPM5-2B + drafter at 100.8 tok/s is the local latency profile we care about. Solid apples-to-apples on one box.