模型训练平台组_机器学习架构工程师（J82075）
Core
Design and implement end-to-end architecture for generative recommendation large models in commercial advertising scenarios, covering pre-training, fine-tuning, inference, and online evaluation.
Role type
Senior Machine Learning Architect (Generative AI/Recommendation Systems)
Builds
High-throughput, high-stability training and inference platforms for generative recommendation models
Domain
Advertising, Recommendation Systems, Generative AI
Deliverable
production ML models
Required skills
Transformer architecture, DeepSpeed, FlashAttention, SMoE, C++, Python, Linux, CUDA, GPU cluster optimization, distributed training, CTR/CVR estimation, sequence modeling, multi-modal large models
Preferred skills
PhD in CS/AI/Math, publications in MLSys/SIGIR/KDD, experience with PaddlePaddle, vLLM, PagedAttention
Technologies
PyTorch, TensorFlow, DeepSpeed, Megatron-LM, FlashAttention, SMoE, vLLM, C++, Python, CUDA
Responsibilities
Design end-to-end model engineering architecture for generative recommendation; optimize training acceleration and inference; collaborate with strategy teams to deploy algorithms to production; track and implement cutting-edge research in recommendation and generative AI
Seniority
Senior, hands-on IC
