AWS Trainium / NKI Kernel Expert
Anyone AI · Remote · Argentina - Fully Remote · Uruguay · Chile · Ecuador - Fully Remote · Portugal · Mexico - Fully Remote · Colombia - Fully Remote · Spain · Brazil ·
COMPENSATION
€•• – ••k
Listed on the employer's job board (Ashby)
- Fully remote
- Part-time contract
- AWS Trainium focus
- NKI kernel expert
- AI workload optimization
Anyone AI seeks engineers with deep expertise in AWS Trainium and the Neuron Kernel Interface (NKI) for a specialized consulting project. You will evaluate kernel development tasks, ensuring they are optimized for Trainium architecture rather than just translated from GPU code.
The work involves reviewing technical implementations for correctness, performance, and idiomatic usage of Trainium-specific patterns like tile-based computation and memory management.
What you'll do
- Evaluate NKI kernel correctness and Trainium-specific development patterns
- Assess CUDA to NKI kernel migrations for quality and efficiency
- Optimize and benchmark Trainium performance for AI workloads
- Analyze memory management across SBUF, PSUM, and HBM
- Review tile-based computation and DMA scheduling strategies
- Verify cross-platform numerical correctness between CUDA/Triton and NKI
- Identify Trainium-specific performance bottlenecks and opportunities
- Provide clear written feedback on kernel implementation quality
What you'll need
- At least 2 years of hands-on NKI kernel development or optimization
- Practical experience with AWS Trainium or Inferentia2 hardware
- Strong grasp of tile-based computation and partition constraints
- Deep understanding of SBUF, PSUM, and HBM memory hierarchy
- Expertise in DMA orchestration and Trainium optimization techniques
- Ability to evaluate and guide CUDA to NKI migration efforts
- Experience profiling and optimizing workloads on Trainium instances
- Understanding of numerical differences between GPU and Trainium backends
- Strong analytical skills for reviewing complex technical implementations
Nice to have
- Familiarity with AWS Neuron SDK or Neuron Compiler
- Background in CUDA or Triton kernel development
- Knowledge of NeuronCore-v2 architecture details
- Experience with FP32, BF16, FP8, and INT8 workloads
- Benchmarking experience on Trn1 or Trn2 instances
- Familiarity with nki.language, jit, or XLA custom calls
- Background in technical evaluation, RLHF, or rubric-based assessment
Stack
- AWS Trainium
- Inferentia2
- NKI
- CUDA
- Triton
- AWS Neuron SDK
- XLA
How you'll work
- Fully remote position
- Part-time engagement
- Project-based consulting role