GenAI Research Engineer
Apple
- Location
- Onsite (Cupertino, California · Seattle, Washington)
- Employment
- Full-time
- Level
- Senior Level
Posted 4 days ago
About the role
Join Apple's Intelligent System Experiences team to advance state-of-the-art generative AI technology behind products like Apple Intelligence, Genmoji, and Image Playground. You will develop and deploy optimization algorithms to make diffusion models faster, smaller, and more efficient for on-device and server deployment.
Skills
Generative AI
Diffusion Models
Quantization
Model Distillation
Kernel Fusion
PyTorch
JAX
Neural Architecture Search
Hardware-aware Architecture
Inference Optimization
Distributed Training
Foundation Models
Pruning
Computer Science
Applied Mathematics
Statistics
Full job details
Do you believe that generative models can fundamentally change how people interact with devices? Are you excited by the possibility of connecting people to unprecedented levels of creativity? Come join the team that turned cutting edge Generative AI research into compelling user experiences through Apple Intelligence. We are part of the Intelligent System Experiences team at Apple and work on the core modeling technology behind Genmoji, Image Playground and Image Wand. From web-scale datasets to distributed training of foundation models, application specific adapters to on-device optimizations, you will have the unique opportunity to advance and ship state-of-the-art technology that impacts Apple users around the world!
We're looking for a Research Engineer to make these models dramatically faster, smaller and cheaper to run on-device and on server. You'll work across the whole efficiency stack: Develop and deploy optimizations algorithms to speed up inference of our production diffusion models (PTQ, QAT, kernel fusion, model distillation). Design hardware-aware architectures that fit within compute and memory limits with minimal loss of quality. Own the process end-to-end from prototyping, evaluation to deployment into the Apple ecosystem.
M.S. or PhD in Computer Science or a related field such as Electrical Engineering, Robotics, Statistics, Applied Mathematics, or equivalent experience. Deep, demonstrated expertise in at least one of: Quantization (low-bit and mixed-precision methods, quantization-aware training). Efficient architecture design, structured pruning or neural architecture search for GenAI or Vision applications. Strong hands-on experience training and fine-tuning diffusion or other large generative models in PyTorch, JAX, or an equivalent framework. Comfortable reasoning about efficiency beyond FLOPs, such as memory, latency and numerical precision, and eager to learn hardware-specific constraints. Excellent cross-functional collaboration skills: You can explain model trade-offs to hardware engineers, hardware constraints to researchers, and integration needs to software teams, in writing and in person.
Publications at CVPR, ICCV, NeurIPS, ICML, ICLR or MLSys on efficient generative models, or significant open-source contributions in this area.
Description
We're looking for a Research Engineer to make these models dramatically faster, smaller and cheaper to run on-device and on server. You'll work across the whole efficiency stack: Develop and deploy optimizations algorithms to speed up inference of our production diffusion models (PTQ, QAT, kernel fusion, model distillation). Design hardware-aware architectures that fit within compute and memory limits with minimal loss of quality. Own the process end-to-end from prototyping, evaluation to deployment into the Apple ecosystem.
Minimum Qualifications
M.S. or PhD in Computer Science or a related field such as Electrical Engineering, Robotics, Statistics, Applied Mathematics, or equivalent experience. Deep, demonstrated expertise in at least one of: Quantization (low-bit and mixed-precision methods, quantization-aware training). Efficient architecture design, structured pruning or neural architecture search for GenAI or Vision applications. Strong hands-on experience training and fine-tuning diffusion or other large generative models in PyTorch, JAX, or an equivalent framework. Comfortable reasoning about efficiency beyond FLOPs, such as memory, latency and numerical precision, and eager to learn hardware-specific constraints. Excellent cross-functional collaboration skills: You can explain model trade-offs to hardware engineers, hardware constraints to researchers, and integration needs to software teams, in writing and in person.
Preferred Qualifications
Publications at CVPR, ICCV, NeurIPS, ICML, ICLR or MLSys on efficient generative models, or significant open-source contributions in this area.