AI Infrastructure Engineer
Unitree Robotics is a Hangzhou-based public robotics company that develops and sells legged and humanoid robots for consumer, education, industrial, and research applications.
Investors
About Unitree Robotics
The brand Unitree Robotics is operated by Yushu Technology Co., Ltd. (宇树科技股份有限公司). Its current portfolio includes quadruped robots, humanoid robots, robotic arms, perception hardware, and associated support and developer resources.
Skills
Candidate Availability
Required and preferred rules are kept separate and reflect the wording in the original posting.
About the Role
You will design AI frameworks for large-scale compute clusters and build distributed training systems for large models. You will optimize GPU utilization, memory use, training throughput, and data-loader efficiency. You will support multimodal large-model and control-model training. You will also build low-latency inference pipelines for real-time robot control using quantization, distillation, and model compilation.
Requirements
- Experience training on GPU clusters with thousands of GPUs
- Knowledge of distributed training mechanisms such as DDP and FSDP
- Experience with DeepSpeed and Megatron-LM
- Ability to define parallelization strategies based on model structures
- End-to-end performance analysis skills for compute, communication, and data-loading bottlenecks
Responsibilities
- Design and build AI frameworks for large-scale compute clusters
- Build and maintain distributed training systems for large-model training
- Optimize GPU utilization, memory consumption, and training throughput
- Optimize data-loader efficiency
- Support training optimization for multimodal large models and control models
- Build low-latency inference pipelines for real-time robot control
- Apply quantization, distillation, and model-compilation techniques to improve inference performance