{
  "operator": "gemm",
  "iterations": 100,
  "input_shape_groups": {
    "M": [4, 8, 16, 32, 64, 128, 256, 512, 1024, 2048, 4096, 8192, 16384, 32768, 65536, 131072],
    "KN": [
      [1024, 1024], 
      [4096, 4096], 
      [8192, 8192], 
      [12288, 12288], 
      
      [16384, 32], 
      [16384, 128], 
      [16384, 1024], 
      
      [32, 16384], 
      [128, 16384],
      [1024, 16384] 
    ]
  },
  "dtype": [
    "float32",
    "bfloat16",
    "float16", 
    "int8"
  ]
}