Tensorflow with TensorRT version compatibility

Description

From this tutorial I installed the tensorflow-GPU 1.15.3 using pip3 command (Not from source) and tensorRT 7.2.1. But when I ran the following commands:

from tensorflow.compiler.tf2tensorrt.wrap_py_utils import get_linked_tensorrt_version
from tensorflow.compiler.tf2tensorrt.wrap_py_utils import get_loaded_tensorrt_version

print(f"Linked TensorRT version {get_linked_tensorrt_version()}")
print(f"Loaded TensorRT version {get_loaded_tensorrt_version()}")

To check the linked tensorRT version. But the reported version was (5, 1, 5) and tensorRT seemed to look for .so.5 lib files in the tensorRT folder, which didn’t make sense at all.

Environment

TensorRT Version: 7.2.1
GPU Type: GTX-1060
Nvidia Driver Version: 450.80.02
CUDA Version: 11.0
CUDNN Version: 8.0.4
Operating System + Version: Ubuntu 18.04 LTS 64-bit
Python Version (if applicable): 3.6.9
TensorFlow Version (if applicable): 1.15.3
PyTorch Version (if applicable): NA
Baremetal or Container (if container which image + tag):

Relevant Files

Please attach or include links to any models, data, files, or scripts necessary to reproduce your issue. (Github repo, Google Drive, Dropbox, etc.)

Steps To Reproduce

Please include:

  • Exact steps/commands to build your repro
  • Exact steps/commands to run your repro
  • Full traceback of errors encountered

Hi @Learner1,
Can you please try doing the setup following the below link

Also the version is determined when building the TensorFlow
If TF build it with a TensorRT 5.1.5, then it will always try to find a .so.5 file.
And it will report 5.1.5
So if you want to use newer version, you will need a rebuild.

Thanks!

An incomplete response!!! The Nvidia docs for trt specify one version whereas tensorflow (pip) linked version is another. The linked doc doesn’t specify how to unlink a trt version or how to build tensorflow with specific tensorrt version. It’s frustrating when despite following all the instructions from Nvidia docs there are still issues. One would expect tensorrt to work with package versions specified in Compatibility section of release notes…

4 Likes