{
  "problem_name": "qknorm_optimization",
  "description": "Optimization problem for high-performance QKNorm using flashinfer",
  "requirements": {
    "cuda_backend": true,
    "gpu_required": true,
    "torch_version": ">=2.0.0",
    "flashinfer_version": "==0.5.0"
  },
  "evaluation": {
    "timeout_seconds": 300,
    "memory_limit_mb": 8192,
    "gpu_memory_limit_mb": 4096
  }
}