大模型训练框架工程师
深圳、合肥、上海
社招
全职
数字技术
本科及以上
1-3 年
职位描述
· 负责大模型分布式训练框架核心组件的方案设计与特性开发 · 研究并优化3D并行(数据/张量/流水线并行)、ZeRO等分布式训练策略,提升千卡级集群扩展效率 · 探索通信优化、计算通信重叠、算子融合等技术,降低分布式训练通信开销 · 参与训练稳定性与容错机制建设,保障长周期预训练任务的稳定运行 · 与推理框架团队协同,打通训练到推理的模型导出与部署链路
职位要求
· 本科及以上学历,计算机、人工智能、数学、信息、电子等相关专业,1年以上相关经验 · 精通Python或C++,具备扎实的工程能力与系统级性能优化功底 · 深入理解Transformer架构与大模型训练全流程(前向/反向/优化器/混合精度等) · 熟悉至少一种主流训练框架(Megatron-LM/DeepSpeed/FSDP等)及其底层实现 加分项(非必需) · 有千卡级以上大规模预训练实战经验,熟悉NCCL/RDMA等高性能通信方案 · 在3D并行、ZeRO优化、序列并行、选择性重计算等方向有深度实践 · 熟悉CUDA/Triton算子开发,有自定义算子融合优化经验 · 有MoE模型训练、长上下文训练等前沿方向落地经验 · 在Megatron-LM/DeepSpeed/Megatron-Deepspeed等开源社区有贡献,或在顶会发表过相关论文
投递

NIO is a pioneer and a leading company in the premium smart electric vehicle market. Founded in November 2014, NIO’s mission is to shape a sustainable and brighter future together. NIO aims to build a community starting with smart electric vehicles to share joy and grow together with users.
NIO designs, develops, jointly manufactures and sells premium smart electric vehicles, driving innovations in next-generation technologies in autonomous driving, digital technologies, electric powertrains and batteries. NIO differentiates itself through its continuous technological breakthroughs and innovations, such as its industry-leading battery swapping technologies, Battery as a Service, or BaaS, as well as its proprietary autonomous driving technologies and Autonomous Driving as a Service, or ADaaS.
NIO’s models for sale include the all-new smart electric flagship SUV ES8, the smart electric flagship coupe SUV EC7, the smart electric mid-large SUV ES7, the smart electric flagship sedan ET7, the all-new smart electric all-round SUV ES6, the all-new smart electric coupe SUV EC6, the smart electric mid-sized sedan ET5, and the smart electric tourer ET5T.