Closed Nik-V9 closed 1 year ago
Can you enable debug logs by doing
with torch_tensorrt.logging.debug():
model = torch_tensorrt.compile(
model_jit,
inputs=[torch_tensorrt.Input(shape=[1, 640, 1024], dtype=torch.half), torch_tensorrt.Input(shape=[1, 640, 1024], dtype=torch.half)],
enabled_precisions={torch.half},
truncate_long_and_double=True
)
and attach the full log here ?
Thanks!
Bug Description
A runtime error is thrown when compiled torch script model is loaded.
The same code works on a docker image with Torch version: 1.10.0a0+0aef44c, Torch-TensorRT version: 1.0.0+55c3bab4, TensorRT version: 8.0.3-1, and CUDA 11.3.
To Reproduce
Code:
Output:
Expected behavior
Model loads successfully.
Environment
conda
,pip
,libtorch
, source): pip (after uninstalling from NGC Docker image nvcr.io/nvidia/pytorch:22.11-py3)Additional context