Research Scientist / Engineer – Training Infrastructure
Posted Oct 8 (45 days ago) · Last updated Nov 22
Posted Oct 8 (45 days ago)
Key facts
Build distributed training systems for multimodal models across thousands of GPUs; remote within the EU.
- Field
- Software engineering
- Languages required
English
Original job ad
About the role
Summarised by JobCrawls from the job ad above.
Requirements
- Extensive experience with distributed PyTorch training and parallelism for foundation models.
- Deep understanding of GPU clusters, networking and storage systems.
- Familiarity with NCCL, MPI and distributed-system optimization.
Responsibilities
- Design, implement and optimize distributed training systems for models running across thousands of GPUs.
- Research and implement FSDP, tensor, pipeline and expert parallelism.
- Build monitoring, visualization and debugging tools for large training runs.
- Optimize training stability, convergence and resource utilization across large clusters.
Skills & technologies
Show 6 more
About Luma AI
| Location | Active listings |
|---|---|
| Remote, EU | 6 |
| London, UK | 2 |
| New York, NY | 2 |
| Redwood City, CA | 2 |
| Remote, US | 2 |
| Seattle, WA | 2 |
| Remote, INT | 1 |
| Remote, UK | 1 |
| Role type | Active listings |
|---|---|
| Performance Optimization Engineer | 2 |
| Distributed Training Engineer | 1 |
| Inference Software Engineer | 1 |
| Reinforcement Learning Infrastructure Engineer | 1 |
| Training Infrastructure Engineer | 1 |
| Role level | Active listings |
|---|---|
| Senior | 4 |
| Expert | 2 |
Luma aims to build unified general intelligence that can generate, understand and operate in the physical world. It believes multimodality is critical to intelligence, with vision as the next step beyond language models.
