大模型训练稳定性和容错系统专家 - Seed Model
Core
Designing and optimizing stability architecture and fault-tolerance mechanisms for ultra-large-scale distributed training clusters to ensure reliable model training.
Role type
Senior IC distributed training stability and fault-tolerance engineer
Builds
Stable, fault-tolerant large-scale distributed training infrastructure for foundation models
Domain
AI / Large Language Models / Distributed Systems
Deliverable
production ML models
Required skills
Distributed training principles, Python/C++/Go, PyTorch, NCCL, RDMA, fault-tolerance mechanisms, automated root cause analysis, cluster stability metrics
Preferred skills
Megatron-LM, DeepSpeed, GPU hardware characteristics, domestic heterogeneous computing, training performance optimization
Technologies
PyTorch, Megatron-LM, DeepSpeed, NCCL, RDMA
Responsibilities
Design and iterate stability architecture for distributed training clusters; Develop and implement fault-tolerance mechanisms with second-level anomaly detection and automatic recovery; Build intelligent root cause analysis systems for training failures; Research and apply frontier technologies in large-scale training stability and optimization.