Model Optimization Engineer

Project

Work done for a product or program.

• Apply structured and unstructured pruning, including 2:4 and block sparsity patterns matched to the hardware's sparse tensor core support.
• Run hardware-aware neural architecture search constrained by latency, memory bandwidth and energy budgets measured on real silicon.
• Implement low-rank factorization, weight sharing and embedding compression for memory-bound layers.
• Characterize the accuracy-latency-energy Pareto frontier for each model family and target device.
• Combine compression techniques with quantization in a single validated pipeline and resolve interaction effects.
• Prototype speculative decoding, early-exit and mixture-of-experts sparsity strategies for LLM workloads.
• Publish reproducible compressed reference checkpoints for the model zoo and customer evaluation.
• Feed measured sparsity utilization data back to architects to justify or retire sparse hardware features.

Function

Work that keeps the discipline running: methods, tools, suppliers and people.

• Design knowledge distillation pipelines that transfer teacher-model quality into smaller student architectures.
• Build fine-tuning and recovery training loops that restore accuracy after aggressive compression.

Project

1. Produce and hand over the role's key deliverables
   • Compressed reference model checkpoints with documented accuracy and latency
   • Pruning and distillation recipe library per model family
   • Hardware-aware NAS search space definitions and search results
   • Accuracy-latency-energy Pareto reports per target device
   • Combined compression-plus-quantization pipeline documentation
2. Develop the softwareGeneral
   • Design to the hardware specification
   • Write and review the code
   • Test on models, emulation or silicon
3. Release and supportGeneral
   • Package and document releases
   • Fix customer-reported defects

Function

1. Maintain the platformGeneral
   • Keep build, test and release tools working
   • Track toolchain and operating system changes
2. Raise qualityGeneral
   • Run code reviews
   • Extend automated tests