Machine Learning Infrastructure Intern

Please note: This position requires candidates to be proficient in using Chinese as the working language.
 

Compensation: This is a paid position. Compensation will be provided in accordance with company policy. Details will be discussed with qualified candidates during the interview process.
 

This role sits at the boundary between model execution and system software. It focuses on bottleneck propagation, resource utilization, and execution efficiency in training and inference.
 

The core question is how model structure, input pipelines, communication patterns, runtime policy, memory behavior, and hardware topology interact.
 

You will

• Analyze end-to-end bottlenecks across input pipelines, host-device transfer, operator execution, GPU kernels, communication, synchronization, and memory behavior

• Study how parallelization strategies, caching policies, batching decisions, and runtime scheduling affect workload efficiency

• Design and implement key modules in training frameworks, inference engines, or serving systems

• Identify sources of GPU cluster underutilization and drive optimizations at the system, framework, or execution layer

• Evaluate candidate mechanisms using real models and production-adjacent workloads

• Generalize local optimizations into reusable infrastructure capabilities
 

You should have

• Strong PyTorch proficiency

• Strong systems fundamentals in operating systems, distributed systems, or computer architecture

• Ability to connect model behavior to system behavior

• Experience in MLSys projects, research, or engineering work
 

Useful but not required

• Python and C++ development experience

• Exposure to GPU architecture or CUDA programming

• Experience profiling or optimizing ML workloads beyond framework-level usage