research/problems/mamba2_scan/resources/submission_spec.json

{
  "problem_name": "mamba2_scan",
  "description": "Triton kernel optimization problem for high-performance Mamba2 Scan computation",
  "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": {
    "L_list": [2048, 4096],
    "D": 512,
    "chunk": 128,
    "BD": 128
  }
}