ProductsBuildersMulti-source78°08:49Chinese contentCohere 开源 Megakernel 推理系统Cohere 开源了比 vLLM 快 1.58 倍的 Megakernel 推理系统,解决了传统引擎在算子边界处的 GPU 空转问题。#Megakernel#Cohere#North Mini Code#vLLM2 个信源在谈Eventshao__meng@shao__meng3 sourcesSourceRead laterReadWorth followingUsefulFollow Megakernel
ModelsBuilders07:52Chinese contentCohere推出融合LLM解码步骤的MegakernelCohere把LLM解码步骤合并成一个内核,大幅提升GPU效率,代码已开源。#Cohere#Megakernel#LLM#GPU官方账号Cohere@cohereSourceRead laterReadWorth followingUsefulFollow Cohere