高性能集合通信库工程师-网络
Core
Design, develop, and optimize a high-performance collective communication library (CCL) for multi-card/multi-machine AI clusters, focusing on RDMA and Scale-up interconnects.
Role type
Senior IC machine-learning infrastructure engineer (collective communication)
Builds
High-speed communication libraries and tools for distributed AI training and inference
Domain
AI infrastructure, distributed systems, high-performance computing
Deliverable
production ML models
Required skills
RDMA technology, collective communication algorithms (Ring, Tree, Halving-Doubling), distributed system architecture, performance tuning, fault diagnosis, Triton
Preferred skills
Communication-computation fusion, NCCL implementation principles, complex scenario adaptation
Technologies
RDMA, Scale-up, Triton, CCL, NCCL
Responsibilities
Design and optimize collective communication libraries for multi-node scenarios; Analyze and resolve performance bottlenecks in AllReduce/All2All operations; Develop fused communication-computation operators to reduce end-to-end latency; Collaborate with training/inference teams to optimize AI model communication efficiency.
