🐛 [Bug] Compilation Error on Citrinet-1024

[gs-olive]

Bug Description

When converting the Citrinet-1024 speech recognition network from TorchScript to Torch-TRT, the following error is encountered:

ERROR: [Torch-TensorRT TorchScript Conversion Context] - 3: %66 : Tensor = aten::sum(%mask2.2, %65, %26, %3779)
# ReduceLayer only supports Float, Half, Int8, and Int32 data types.
Segmentation fault (core dumped)

To Reproduce

Steps to reproduce the behavior:

1. Run torch_tensorrt.compile with Citrinet-1024 model as input, using fp32 precision.
2. Choose fixed input sizes of [1, 80, 256] and [1,] and enable truncate_long_and_double with 8 GB workspace.

Expected behavior

Model should successfully compile to Torch-TRT. Specifically, internal (non-user-provided) type-casting issues should not cause errors.

Environment

• Torch-TensorRT Version: 1.3.0a0+1361028c
• 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