后端开发工程师（交易与广告） - 穿山甲
Core
Optimizing training and serving efficiency for billion-parameter deep learning models across the recall, coarse, and fine-grained recommendation chain.
Role type
Senior Machine Learning Infrastructure Engineer (Model Efficiency & Training Systems)
Builds
High-performance training systems and optimized model serving pipelines for user growth and ad monetization.
Domain
Internet Advertising & Recommendation Systems
Deliverable
production ML models
Required skills
Deep learning frameworks (PyTorch), distributed training, model quantization/compression, mixed-precision training, CUDA optimization, system-level tuning (GPU/TPU, memory, communication), C++/Python/Go/Java
Preferred skills
LLM/long-sequence model optimization, search/ad/recommendation scenario experience, model distillation
Technologies
PyTorch, CUDA, GPU/TPU, C++, Python, Go, Java
Responsibilities
Optimize training speed, model size, and serving latency; implement quantization and mixed-precision techniques; tune data preprocessing and gradient synchronization; develop custom training system features; collaborate with algorithm teams for rapid iteration.