🐛 [Bug] Encountered bug when converting deepset/roberta-base-squad2 Hugging Face Network

[gs-olive]

Bug Description

When converting the deepset/roberta-base-squad2 Hugging Face network (https://huggingface.co/deepset/roberta-base-squad2) from TorchScript to Torch-TRT, the following error is encountered:

RuntimeError: [Error thrown at core/conversion/conversion.cpp:428] Expected eval_list->elements().size() == n->outputs().size() to be true but got false
Size of evaluated results: 3 and node outputs size: 2 must match.

To Reproduce

Steps to reproduce the behavior:

1. Run torch_tensorrt.compile with deepset/roberta-base-squad2 model as input, using fp32 precision.
2. Choose fixed input size of [1, 128] and enable truncate_long_and_double with 8 GB workspace.
3. Pass in model keyword args to disable attention and hidden state outputs

Expected behavior

Model should successfully compile to Torch-TRT. Specifically, the size of evaluated results and node outputs should be matching

Environment

• Torch-TensorRT Version: 1.3.0a0+b70c9132
• PyTorch Version: 1.13.0.dev20220921+cu116
• CPU Architecture: Intel Xeon CPU
• OS: Ubuntu 20.04
• How you installed PyTorch: pip
• Build command you used: python setup.py develop
• Are you using local sources or building from archives: local
• Python version: 3.8.13
• CUDA version: 11.6

Additional context

Bug was reproduced on PyTorch 22.08 and 22.09 containers