Research Scientist / Engineer – Training Infrastructure
Luma · Remote · Remote, EU ·
COMPENSATION
€•• – ••k
Listed on the employer's job board (Ashby)
- Fully remote
- PyTorch + CUDA
- Distributed systems
- Thousands of GPUs
- Senior level
You will design and optimize the distributed systems that power Luma's foundation model training. This position focuses on solving complex challenges in parallelism, stability, and resource utilization across massive GPU clusters, enabling researchers to innovate efficiently.
What you'll do
- Design and optimize distributed training systems for models on thousands of GPUs
- Implement advanced parallelization strategies such as FSDP and Tensor Parallel
- Build monitoring, visualization, and debugging tools for large-scale training
- Improve training stability, convergence, and resource utilization across clusters
What you'll need
- Extensive experience with distributed PyTorch training and parallelisms
- Deep understanding of GPU clusters, networking, and storage systems
- Familiarity with communication libraries like NCCL and MPI
- Proven track record in distributed-system optimization for foundation models
Nice to have
- Strong Linux systems administration and scripting skills
- Experience managing training runs across more than 100 GPUs
- Background in containerization, orchestration, and cloud infrastructure
Stack
- PyTorch
- CUDA
- FSDP
- NCCL
- MPI
- Linux