Put your Neuron Kernel Interface experience to work as a reviewer. You judge NKI kernel tasks for correctness, performance on Trainium hardware and faithful CUDA to NKI migration, then write clear feedback against a rubric.
What you will do
- Review NKI kernel development tasks for quality, correctness and fit with Trainium hardware.
- Check how faithfully CUDA kernels were migrated to NKI.
- Judge Trainium-specific performance optimization and cross-platform numerical correctness.
- Write clear, rubric-based feedback on each task.
What you need
- 2+ years of hands-on work developing or optimizing kernels with NKI for AWS Trainium or Inferentia2.
- A strong grasp of NKI patterns: tile-based computation, SBUF/PSUM/HBM memory management, partition-dimension limits and DMA orchestration.
- Experience assessing the quality of CUDA to NKI migrations.
- Familiarity with Trainium performance profiling, such as NeuronCore pipeline use, tensor-engine throughput and memory-bandwidth bottlenecks.
- Experience setting or checking numerical-correctness standards across GPU and Trainium (accumulation order, rounding, mixed precision).
A plus: work with the AWS Neuron SDK or Neuron Compiler internals, contributions to NKI kernel libraries, CUDA or Triton kernel work, knowledge of NeuronCore-v2 architecture and supported data types, and benchmarking ML training on Trn1 or Trn2 instances.
How pay works
The listing gives an hourly rate of $70-90/hr.
Apply (opens in a new tab)