🐛 [Bug] Failed to use stoi to get the engine ID in TRTEngine function

[bowang007]

Bug Description

Got this error:

INFO: [Torch-TensorRT - Debug Build] - [MemUsageChange] TensorRT-managed allocation in IExecutionContext creation: CPU +0, GPU +132, now: CPU 0, GPU 540 (MiB)
terminate called after throwing an instance of 'std::invalid_argument'
  what():  stoi
Aborted (core dumped)

Debugged and found that it fails at this line: https://github.com/NVIDIA/Torch-TensorRT/blob/c95229144432d96f4bdaa71fb1d242242d42bc29/core/runtime/TRTEngine.cpp#L63

To Reproduce

MaskRCNN model from detectron2.

Expected behavior

should figure out why we cannot find a number from this engine.

[narendasan]

@bowang007 I have a fix for this

[bowang007]

could you please share more details? @narendasan

[narendasan]

We still have the issue where TensorRT is changing the names of our input bindings so leaving this open to track this issue.

[bowang007]

closing per #1167