Role OverviewMercor is seeking a GPU Kernel Developer - AI Trainer to evaluate the quality and correctness of GPU/accelerator kernel development tasks for training and evaluating AI models. The ideal candidate will have 3+ years of experience with CUDA, Triton, NKI, or Pallas (JAX) and a strong understanding of numerical-correctness criteria for kernels.
What You Will Do
Evaluate kernel development tasks, assess numerical correctness, performance-benchmarking fairness, and task scoping, review compilation and runtime validity, and collaborate with AI research teams to improve model training and evaluation processes.
Why It Might Be a Fit
The successful candidate will work independently and asynchronously to meet deadlines and enhance AI model performance. Experience with performance profiling and benchmarking, familiarity with common compilation and runtime failure modes, and experience with at least three kernel task types are preferred.
Requirements
- 3+ years of experience with CUDA, Triton, NKI, or Pallas (JAX)
- Strong understanding of numerical-correctness criteria for kernels
- Experience with performance profiling and benchmarking (e.g., nsight, ncu)
- Familiarity with common compilation and runtime failure modes
- Experience with at least three kernel task types
]]>