Data Scientist - Model Optimization

Responsibilities

• Design statistically rigorous experiments to compare PTQ, QAT, and mixed-precision schemes on vision, language, and multimodal models.
• Implement custom quantization algorithms from scratch, adapting existing techniques or developing novel approaches to match Chimera GPNPU's unique architectural features and numerical formats.
• Build calibration datasets; develop Python notebooks/dashboards to track accuracy, latency, power, and memory trade-offs.
• Perform layer-level error analysis to guide numerical-format choices.
• Partner with compiler team to convert your findings into turnkey SDK flows and reference configs.
• Publish internal white papers, external benchmarks, and present results to customers and at industry events.
• Monitor academic literature in compression and efficient inference; translate promising ideas into reproducible prototypes.

Requirements

• M.S./Ph.D. in CS, EE, Applied Math, or similar, with 5+ years in ML model optimization or data-science-driven research.
• Deep grasp of fixed-point arithmetic, quantization theory, numerical analysis, and statistical calibration.
• Strong ability to implement quantization algorithms from first principles, not just use existing frameworks.
• Fluent in Python, PyTorch or TensorFlow, NumPy/Pandas/SciPy, and data-viz tools (Matplotlib/Plotly).
• Experience implementing custom quantizers and understanding their interaction with hardware constraints (bit-width, format, operations).
• Hands-on with at least one quantization toolkit (PyTorch FX/PTQ/QAT, TF-Lite, ONNX-Runtime, TVM, MLIR Quant) and ability to extend them.
• Working knowledge of CNNs, Transformers, and DNN architectures.

Bonus

• Experience with custom hardware accelerators, DSPs, or neural processing units.