Responsibilities
- Develop a full-cycle machine learning framework tailored for 3D data, covering pretraining, fine-tuning, and inference stages
- Collaborate with research teams to jointly develop next-generation 3D and spatial artificial intelligence technologies
- Implement and troubleshoot advanced PyTorch-based systems to achieve optimal parallelization and performance
- Create streamlined and user-friendly training pipelines for proprietary foundational models
- Diagnose performance constraints and enhance throughput for distributed model training across large-scale GPU clusters
- Develop and support custom 3D operators using Triton or CUDA for specialized computations
- Build and sustain innovative data loading frameworks and supporting libraries
- Design high-efficiency inference services that manage multi-stage model workflows
- Improve model performance using techniques such as compilation, operator fusion, and quantization
Work Arrangement
On-site — Sunnyvale
Other
- Must be able to communicate information and ideas clearly to ensure understanding
- Must be able to observe fine details while working at close range
- Must be able to meet project deadlines and work under time constraints
