Institute of Foundation Models
Machine Learning Engineer — Pre-training (LLM)
$150K–$450K
Sunnyvale, CA · On-site · Full-time · Posted Oct 5, 2026
- Experience
- 5+ years in ML systems, infrastructure, or distributed training; large-scale pre-training experience is required.
About the job
Build and scale distributed ML infrastructure for large-scale LLM pre-training across multi-node, multi-GPU clusters.
What you'll do
- Extend distributed training frameworks and implement distributed optimizers from mathematical specifications.
- Build launch, experiment-tracking, monitoring, data-loading, and checkpoint/restart systems for training workloads.
- Improve the reliability, maintainability, and performance of ML infrastructure using production-quality code and tests.
What they ask for
- Experience modifying distributed ML frameworks such as DeepSpeed, FSDP, FairScale, or Horovod.
- Strong software engineering fundamentals, including Python, systems design, and testing.
- Multi-node experience and debugging skills, including tools such as Slurm, Kubernetes, Ray, NCCL, or GLOO.
- Experience with ML infrastructure or distributed inference optimization, large-scale ML workloads, and large-scale pre-training.
Nice to have
- Exposure to mixed-precision training, performance profiling, kernel fusion, or memory optimization.
- CUDA or Triton kernel experience, open-source contributions, or published research.
Pay
USD 150,000–450,000 per-year-salary (base compensation)
Tools
DeepSpeed, FSDP, FairScale, Horovod, Python, PyTorch, JAX, Slurm
AI focus
Supports large-scale foundation-model pre-training by developing and scaling distributed training infrastructure.
Model-written summary · the original posting is authoritative