Research Engineer, Large-Scale AI Training Infra

ByteDance; San Jose (CA); On-site; USD 254,000 - 480,000; Full time

Job summary

ByteDance Seed in San Jose seeks a researcher/engineer to advance distributed training for foundation models. You will contribute to pretraining efficiency, multimodal models, and high-performance inference in a cutting-edge infrastructure team.

You will work on parallelism strategies, cross-layer optimizations, and production-grade ML systems, collaborating with researchers and engineers globally. A strong background in GPU kernels and large-scale ML training is essential.

Qualifications

• Bachelor's degree or higher in CS/SE or related field with 2+ years in large-scale model training or distributed systems.
• Strong understanding of large-scale training of LLMs and multimodal models.
• Experience with parallelism strategies on large GPU clusters and GPU kernel optimization.
• Algorithm–system co-design for training efficiency, scalability, and reliability.

Responsibilities

• Conduct research and engineering on large-scale infrastructure for efficient pretraining of foundation models.
• Design and optimize distributed training strategies and throughput on large GPU clusters.
• Analyze performance bottlenecks in exascale training systems and propose data-driven optimizations.
• Bridge research and production by translating ideas into scalable infrastructure solutions.

Skills

Large-scale model training

Distributed training systems

GPU kernel optimization

Parallelism strategies

Cross-layer optimization

Programming skills

Education

Bachelor's degree or above in Computer Science/Software Engineering

Tools

Megatron-LM

DeepSpeed

PyTorch FSDP

NCCL

CUDA

Triton

Job description

ByteDance Seed in San Jose seeks a researcher/engineer to advance distributed training for foundation models. You will contribute to pretraining efficiency, multimodal models, and high-performance inference in a cutting-edge infrastructure team.

You will work on parallelism strategies, cross-layer optimizations, and production-grade ML systems, collaborating with researchers and engineers globally. A strong background in GPU kernels and large-scale ML training is essential.