2月4日,腾讯混元AI Infra团队正式推出开源生产级高性能LLM推理核心算子库HPC-Ops。腾讯混元称,在真实场景下,基于HPC-Ops,混元模型推理QPM提升30%,DeepSeek模型QPM提升17%。同时,在单算子性能方面,HPC-Ops实现Attention相比FlashInfer/FlashAttention最高提升2.22倍;GroupGEMM相比DeepGEMM最高提升1.88倍;FusedMoE相比TensorRT-LLM最高提升1.49倍。
特别声明:以上内容(如有图片或视频亦包括在内)为自媒体平台“网易号”用户上传并发布,本平台仅提供信息存储服务。
Notice: The content above (including the pictures and videos if any) is uploaded and posted by a user of NetEase Hao, which is a social media platform and only provides information storage services.