The mission of our AML team is to push the next-generation AI infrastructure and recommendation platform for the ads ranking, search ranking, live & ecom ranking in our company. We also drive substantial impact on core businesses of the company. Currently, we are looking for Machine Learning Engineer in AI Compiler Optimization to join our team to support and advance that mission.
Responsibilities:
- Responsible for building and implementing the compilation optimization system for the recommendation machine learning engine. Design and implement full-stack optimization solutions at the graph, operator, and memory levels specifically for recommendation model scenarios, including but not limited to graph-operator fusion and automatic operator generation, to maximize hardware computing limits.
- Collaborate closely with hardware and algorithm teams to carry out hardware-software co-design. Optimize compilation strategies based on hardware characteristics to improve the efficiency of hardware-software synergy.
- Responsible for the compilation adaptation of recommendation models from the PyTorch framework to the engine. Optimize the entire process of model import, conversion, and code generation to simplify the model deployment process and enhance development efficiency.
Minimum Qualifications:
- Proficient in one of the mainstream AI compiler frameworks (e.g., Triton, MLIR, TVM), with practical project experience in customized compilation optimization and Pass development based on the framework.
- Experience in GPU/NPU compilation optimization, mastering core techniques such as loop optimization, memory optimization, and operator optimization, with the ability to independently perform performance bottleneck analysis and technical optimization.
- Familiar with common model structures and compilation adaptation logic of deep learning frameworks such as PyTorch and TensorFlow, capable of designing targeted optimization solutions.
Preferred Qualifications:
- Familiar with the architecture design of recommendation machine learning engines. Solid implementation experience in compilation optimization and low-latency inference optimization for large-scale recommendation systems, with the ability to handle compilation optimization needs in high-concurrency scenarios.
- Experience contributing to open-source AI compiler projects (e.g., TVM, MLIR), or possess technical expertise in the compilation adaptation of large models for recommendation scenarios and the automatic generation of sparse operators.
