Migrating IEngineInspector Usage from TensorRT 10.x to 11.x

If your application parses the JSON output of IEngineInspector::getLayerInformation() or IEngineInspector::getEngineInformation(), the output schema has changed in TensorRT 11.x. Update your parsing logic as described below.

For general Engine Inspector usage and API details, refer to the Engine Inspector documentation.

Bindings Field Replaced by I/O Tensors

The Bindings field, which was a flat list of I/O tensor names as strings, has been removed. It is replaced by the new I/O Tensors field, which is a list of JSON objects. Each object contains structured information about the tensor, including its name, I/O mode, data type, dimensions, memory location, and whether it is a shape inference I/O.

Before (TensorRT 10.x)

{
  "Bindings": [
    "input_0",
    "output_0"
  ]
}

After (TensorRT 11.x)

{
   "I/O Tensors": [
     {
       "Name": "input_0",
       "IOMode": "Input",
       "DataType": "Float",
       "Dimensions": [1, 3, 224, 224],
       "Location": "Device",
       "IsShapeInferenceIO": false,
       "ProfileInfo": [
       {
         "MinShape": [1,3,224,224],
         "OptShape": [1,3,224,224],
         "MaxShape": [1,3,224,224],
         "Format": "Row major linear FP32 format (kLINEAR)"
       }]
     },
     {
       "Name": "output_0",
       "IOMode": "Output",
       "DataType": "Float",
       "Dimensions": [1, 1000],
       "Location": "Device",
       "IsShapeInferenceIO": false,
       "ProfileInfo": [
       {
         "Format": "Row major linear FP32 format (kLINEAR)"
       }]
     }
   ]
 }

Format / DataType Field Split

In layer-level information, the combined Format/DataType field has been removed and replaced with two separate fields: Format and DataType. This allows you to read the data type directly without parsing the format string.

TensorRT 10.x Output (Before)

{
  "Name": "Conv_0",
  "Inputs": [
    {
      "Name": "input_0",
      "Format/DataType": "Row major linear FP32"
    }
  ]
}

TensorRT 11.x Output (After)

{
   "Name": "Conv_0",
   "Inputs": [
     {
       "Name": "input_0",
       "Datatype": "Float",
       "Format": "(Linear) Row major linear format"
     }
   ]
 }