TikTok
Research Scientist, (Privacy-Preserving Large-Scale Model Training & Architecture Optimization)
San Jose, California · Full-time · Posted Oct 28, 2026
About the job
Designs and optimizes large-scale training systems and architectures for diffusion-based and unified multimodal foundation models in privacy-sensitive production environments.
What you'll do
- Design distributed training architectures and optimize GPU performance for large-scale generative models.
- Build fault-tolerant training systems with failure detection, recovery, checkpointing, and improved utilization.
- Optimize diffusion and unified generation-understanding model training, collaborating on quality, efficiency, and stability tradeoffs.
What they ask for
- Strong background in large-scale deep learning systems and distributed training.
- Hands-on GPU optimization experience, including memory management, communication/computation overlap, and performance profiling.
- Experience training diffusion models, DiT-style architectures, or large foundation models at scale.
- Proficiency in PyTorch and modern distributed training stacks; solid understanding of parallelism strategies and training robustness.
Nice to have
- Experience with privacy-preserving ML, sensitive-data training, or regulated environments.
- Familiarity with fault-tolerant systems, production GPU orchestration, unified multimodal models, low-level performance work, or infrastructure supporting thousands of GPUs.
Tools
PyTorch, DiT, Rectified Flow, DP, TP, PP, ZeRO, FSDP
AI focus
Develop and optimize diffusion-based and unified multimodal generative foundation models, including privacy-aware training pipelines.
Model-written summary · the original posting is authoritative