{
  "context": "This evaluation assesses how effectively the engineer uses CuPy's JIT compilation capabilities with the @rawkernel decorator to implement custom GPU kernels for element-wise vector operations. The focus is entirely on proper usage of the @rawkernel decorator and related JIT compilation features.",
  "type": "weighted_checklist",
  "checklist": [
    {
      "name": "@rawkernel decorator usage",
      "description": "Uses the @rawkernel decorator from cupy to define at least one custom GPU kernel function. The decorator should be properly imported and applied to Python functions that will be JIT-compiled to CUDA code.",
      "max_score": 30
    },
    {
      "name": "Kernel function implementation",
      "description": "Implements kernel functions with proper CUDA-style syntax that can be compiled by the JIT compiler. Functions should use appropriate indexing (e.g., thread and block indices) and perform the required mathematical operations.",
      "max_score": 25
    },
    {
      "name": "Type annotations",
      "description": "Provides appropriate type hints or annotations for kernel parameters to enable type inference during JIT compilation. This may include using CuPy's type system or Python type hints compatible with the JIT compiler.",
      "max_score": 15
    },
    {
      "name": "Kernel invocation",
      "description": "Correctly invokes the JIT-compiled kernels with proper grid and block dimensions. Should demonstrate understanding of how to launch kernels with appropriate thread configuration.",
      "max_score": 20
    },
    {
      "name": "Multiple operations",
      "description": "Uses @rawkernel or JIT compilation for multiple different operations (at least 3 of the 4 required functions). Demonstrates ability to apply the technique across different mathematical operations rather than just a single use case.",
      "max_score": 10
    }
  ]
}