跳到正文
Cohere· @cohere · X·· 2026-09-09AI 评分51
AI 导读

Cohere 发布首个围绕 decode megakernel 构建的完整 LLM 文本生成服务系统,性能最高比 vLLM 快 1.58x。该系统为 North Mini Code 打造,完全开源。

正文 · 原文

Introducing the next evolution in LLM text generation: the first fully-fledged serving system built around a decode megakernel.

Delivering up to 1.58x faster performance than vLLM. Built for North Mini Code, completely open-source.

来源:Cohere · x.com