AI Systems, Model Optimization

What you'll do

• Model compute, memory, and energy trade-offs across models and hardware configurations.
• Map and optimize AI models for hardware constraints using techniques such as quantization, sparsity, pruning, and distillation.
• Develop hardware-aware training and optimize GPU kernels; profile and debug ML code to address performance bottlenecks.

What they ask for

• MS/PhD or equivalent research/project experience in a quantitative field such as AI/Machine Learning, Computer Science, Physics, Electrical Engineering, or Applied Math.
• Practical understanding of the modern AI/ML stack and optimized execution on modern GPU systems, including performance profiling and bottleneck resolution.
• Ability to map modern AI architectures to system performance and apply efficiency techniques such as sparsity, quantization, and distillation.
• Deep PyTorch experience, including internals, torch.compile, and DDP/FSDP libraries.

Nice to have

• Experience with production-grade training frameworks such as Megatron-LM or DeepSpeed and distributed training at scale.
• Experience with unconventional computing co-design or advanced approximation/compression techniques, including noise-aware or physics-constrained training.

Tools

CUDA, Triton, CUTLASS, PyTorch, torch.compile, DDP, FSDP, Megatron-LM

AI focus

Optimize AI model training and execution for analog compute substrates, including hardware-aware, quantization-aware, noise-aware, and sparsification techniques.