High-Performance ML Backend Engineer

ByteDance; Singapore; Presencial; SGD 120 000 - 180 000; Tempo integral

Resumo da oferta

ByteDance's Data AML team invites graduates to join in 2027, offering opportunities to pursue bold ideas and grow within a fast-evolving ML mid-platform. You will tackle complex challenges and contribute to the training and inference systems powering Douyin, Jinri Toutiao, and Xigua Video.

You will optimize the computational performance of mid-platform models, design high-performance kernel libraries, and advance the model compilation system across graph optimization, fusion, and scheduling.

Qualificações

• Bachelor’s or Master’s degree in computing or related discipline.
• Familiar with model compilation stacks (e.g., TVM, MLIR, XLA) and optimization.
• Proficient in C/C++ development with kernel development experience.
• Understanding of deep learning frameworks and computational graphs.

Responsabilidades

• Optimize computational performance of ByteDance's mid-platform models across inference/training bottlenecks.
• Lead design and development of high-performance kernel libraries for general-purpose and business-customized kernels.
• Advance model compilation technology including graph optimization, kernel fusion, scheduling, and code generation.
• Collaborate with business and algorithm teams to diagnose performance issues and implement full-stack optimizations.

Conhecimentos

C/C++ development

GPU/CPU architectures

Kernel development

Model compilation optimization

Independent problem solving

Formação académica

Bachelor's or Master's in computing or related

Ferramentas

TVM

MLIR

XLA

TensorFlow

PyTorch

OneFlow

Descrição da oferta de emprego

ByteDance's Data AML team invites graduates to join in 2027, offering opportunities to pursue bold ideas and grow within a fast-evolving ML mid-platform. You will tackle complex challenges and contribute to the training and inference systems powering Douyin, Jinri Toutiao, and Xigua Video.

You will optimize the computational performance of mid-platform models, design high-performance kernel libraries, and advance the model compilation system across graph optimization, fusion, and scheduling.