Remote anywhere in Canada, this machine learning systems engineer benchmarks and tunes production LLM serving across AWS Trainium2 and Trainium3. Python and PyTorch skills are central.Capgemini Engineering seeks a hands-on engineer focused on accelerator performance and customer-ready inference systems. You will compare Trainium with leading GPUs, measure latency-throughput frontiers, and optimize price-performance under representative loads.
The work includes Qwen-family models, vLLM, AWS Neuron, long contexts, and multi-chip serving.Key Responsibilities:
Benchmark workloads across Trainium generations and leading GPUsControl prompts, concurrency, cache state,
and warm-upDeliver reproducible price-performance comparisons aligned with service objectivesApply architecture-specific optimizations to long-context workloadsSupport workshops enabling independent operation and tuningRequirements:
Five-plus years of machine learning or systems engineering experienceProduction inference or serving experience for two-plus yearsHands-on AWS Trainium, Inferentia, or AWS Neuron experience preferredExperience profiling accelerator workloads at framework, compiler, or kernel levelStrong written and verbal communication for technical audiencesBring rigorous performance experimentation and accelerator migration expertise to AWS Trainium inference engineering at Capgemini Engineering.