ByteDance
Research Engineer (Seed Model - Infra - Training Performance & Core Kernels)
San Jose, California · Full-time · Posted Oct 28, 2026
- Education
- Bachelor's degree or above in Computer Science, Electrical Engineering, or a related field.
About the job
Designs and optimizes high-performance training kernels for large-scale AI foundation models and integrates them into production training frameworks.
What you'll do
- Design and optimize core kernels, including Attention, GEMM, and MoE, for forward and backward passes.
- Profile and model hardware performance to identify bottlenecks and guide kernel-level optimizations.
- Implement low-precision computation, validate numerical correctness and training stability, and optimize kernels for new hardware.
What they ask for
- Bachelor's degree or above in Computer Science, Electrical Engineering, or a related field.
- Strong proficiency in C/C++ and Python, with foundations in computer architecture and parallel programming.
- Hands-on experience implementing and optimizing high-performance GPU kernels using at least one of CUDA / CUTLASS, Triton, or HIP.
- Ability to identify compute, memory, and communication bottlenecks and support optimization decisions with measured data.
Nice to have
- Experience optimizing training kernels, including backward passes and integration with frameworks such as Megatron.
- Experience with low-precision kernels, cross-generation hardware porting, Triton compiler internals, or ML compiler backends.
Tools
C/C++, Python, CUDA, CUTLASS, Triton, HIP, Megatron
AI focus
Develops and optimizes training infrastructure and kernels for AI foundation models.
Model-written summary · the original posting is authoritative