Data Scientist - Model Optimization
Core
Research, prototype, and implement novel quantization algorithms tailored to Quadric's custom GPNPU architecture to maximize performance on edge and endpoint devices.
Role type
Senior IC data scientist specializing in model optimization and quantization
Builds
Custom low-precision methods and SDK flows for the Chimera GPNPU
Domain
Edge computing, neural network inference, hardware acceleration
Deliverable
production ML models
Required skills
Fixed-point arithmetic, quantization theory, numerical analysis, statistical calibration, Python, PyTorch/TensorFlow, NumPy/Pandas/SciPy, data visualization, CNNs/Transformers/DNNs
Preferred skills
Custom hardware accelerators, DSPs, neural processing units
Technologies
PyTorch FX/PTQ/QAT, TF-Lite, ONNX-Runtime, TVM, MLIR Quant
Responsibilities
Design statistically rigorous experiments comparing PTQ, QAT, and mixed-precision schemes; Implement custom quantization algorithms from scratch; Build calibration datasets and dashboards for accuracy/latency/power/memory trade-offs; Perform layer-level error analysis; Partner with compiler team to convert findings into SDK flows; Publish white papers and benchmarks; Monitor academic literature and translate ideas into prototypes
Seniority
Senior, hands-on IC
