🐛 [Bug] Segmentation fault when using int8 precision

[AhmetHamzaEmra]

Bug Description

tensorrt cannot work with int8 precision when using on NanoGPT like (bark text-to-audio) model.

To Reproduce

Steps to reproduce the behavior:

inp1, inp2 = get_sample_input()

inp1 = torch.unsqueeze(inp1, 0)

traced_model = torch.jit.trace(model, example_inputs=[inp1, inp2])

batch_size = 1

trt_model = torch_tensorrt.compile(
    traced_model,
    inputs = [  torch_tensorrt.Input((batch_size,1), dtype=torch.long), 
                torch_tensorrt.Input((batch_size, 1024, 8), dtype=torch.long)],
    enabled_precisions = { torch.int8},
    workspace_size=20000000000,
    truncate_long_and_double = True
)

and I am getting following error:

Segmentation fault

Expected behavior

When I run the same code on float32 or half precion everyting works fine. only happens when its int8

Environment

Build information about Torch-TensorRT can be found by turning on debug messages

• Torch-TensorRT Version (e.g. 1.0.0): '1.4.0'
• PyTorch Version (e.g. 1.0): 2.0.1+cu117
• CPU Architecture: x64
• OS (e.g., Linux): WSL
• How you installed PyTorch (conda, pip, libtorch, source): pip
• Build command you used (if compiling from source):
• Are you using local sources or building from archives: Local
• Python version: Python 3.10.6
• CUDA version: 11.5
• GPU models and configuration: 4090
• Any other relevant information:

Additional context

[SongDabao]

I have the same issue, any update in this?

[AhmetHamzaEmra]

I am waiting for an update too 😅

[github-actions]

This issue has not seen activity for 90 days, Remove stale label or comment or this will be closed in 10 days

[AhmetHamzaEmra]

Why are you guys keep closing the issue? We are still waitign for an update ?????