manifest.json

{
  "kind": "benchmark",
  "num_problems": 87,
  "input_tensor_spec": "utils/input_output_tensors.py",
  "eval_defaults": {
    "world_size": 8,
    "m": 1024,
    "n": 1024,
    "dtype": "bfloat16",
    "trials": 5,
    "warmup_iters": 500,
    "profiling_iters": 100
  }
}