research/problems/group_gemm/resources/submission_spec.json

{
  "problem_name": "group_gemm",
  "description": "Triton kernel optimization problem for high-performance batched matrix multiplication (BMM)",
  "requirements": {
    "cuda_backend": true,
    "gpu_required": true,
    "triton_version": ">=2.1.0",
    "torch_version": ">=2.0.0"
  },
  "evaluation": {
    "timeout_seconds": 300,
    "memory_limit_mb": 8192,
    "gpu_memory_limit_mb": 4096
  },
  "metadata": {
    "B_list": [64, 256, 1024],
    "M": 64,
    "N": 64,
    "K": 64
  }
}