Trainium NKI Kernel Expert at Mercor

Trainium NKI Kernel Expert | $70–90/hr | Remote (US)

Join a cutting-edge AI evaluation project where your deep expertise in Neuron Kernel Interface (NKI) development will directly shape the quality of training infrastructure for a frontier AI lab. In this role, you will assess, critique, and provide rubric-based written feedback on NKI kernel development tasks — evaluating CUDA→NKI migration fidelity, Trainium-specific performance optimizations, and cross-platform numerical correctness.

What You'll Do

• Evaluate the quality, correctness, and hardware-appropriateness of NKI development tasks used to train and benchmark frontier AI models
• Assess CUDA→NKI migration fidelity, ensuring idiomatic use of Trainium-native patterns
• Review Trainium-specific performance optimization quality, including NeuronCore pipeline utilization and memory-bandwidth efficiency
• Define and apply cross-platform numerical-correctness standards (GPU vs. Trainium accumulation order, rounding behavior, mixed-precision semantics)
• Deliver clear, structured, rubric-based written feedback to guide task improvement

Basic Qualifications

• 2+ years of hands-on experience developing or optimizing kernels using the Neuron Kernel Interface (NKI) targeting AWS Trainium/Inferentia2 hardware
• Strong understanding of NKI-specific development patterns: tile-based computation, SBUF/PSUM/HBM memory-hierarchy management, partition-dimension constraints, and DMA orchestration
• Demonstrated experience assessing CUDA→NKI migration quality
• Familiarity with Trainium-specific performance profiling tools and metrics
• Experience evaluating cross-platform numerical-correctness standards across GPU and Trainium architectures

Preferred Qualifications

• Direct experience with AWS Neuron SDK, Neuron Compiler internals, or contributions to NKI kernel libraries
• Prior CUDA or Triton kernel development experience
• Familiarity with Trainium hardware specifications, including NeuronCore-v2 architecture and supported data types (FP32/BF16/FP8/INT8)
• Experience benchmarking ML training workloads on Trn1/Trn2 instances

This is a fully remote, hourly contractor engagement open to candidates based in the United States.