Data Scientist - Model Optimization

Quadric has developed an innovative general purpose neural processing unit (GPNPU) architecture. Our co-optimized software and hardware are designed to run neural network (NN) inference workloads across various edge and endpoint devices, from battery-operated smart sensors to high-performance automotive and autonomous vehicle systems. Unlike other NPUs or neural network accelerators that can only accelerate parts of a machine learning graph, the Quadric GPNPU executes both NN graph code and conventional C++ DSP and control code.

Our Values :

Integrity, Humility, Happiness

Our Expectations :

Initiative, Collaboration, Completion

Role :

You will join the data science team focused on model optimization. Your work will involve researching, prototyping, and validating low-precision techniques to make neural networks more efficient on the ChimeraGPNPU. Your analyses will influence the quantization recipes in the ChimeraSDK and guide future hardware features.

Responsibilities :

• Design rigorous experiments to compare PTQ, QAT, pruning, and mixed-precision schemes on vision, language, and multimodal models.
• Build calibration datasets and develop Python notebooks or dashboards to track accuracy, latency, power, and memory trade-offs.
• Perform layer- and token-level error analysis to inform numerical format choices.
• Collaborate with the compiler team to translate findings into SDK flows and reference configurations.
• Publish internal whitepapers, external benchmarks, and present results at industry events and to customers.
• Stay updated with academic literature on compression and efficient inference; translate promising ideas into reproducible prototypes.

Qualifications :

Benefits :

Founded in 2016 and based in downtown Burlingame, California, Quadric is building the world’s first supercomputer for real-time edge device needs. Our goal is to empower developers across industries to create innovative technologies today. The company was co-founded by technologists from MIT and Carnegie Mellon, previously involved in Bitcoin computing at 21.

#J-18808-Ljbffr