tensorflow
Support for TensorRT 8.0
In order to work with the TensorRT Api we have set up an Ubuntu 20.04 installation with TensorRT 8.0 which is the only version available for Ubuntu 20.04. When executing the API with an EfficientDetD0 I run into the Error "Could not load dynamic library 'libnvinfer.so.7'. I assume it is searching for the TensorRT 7.x installation which I do not have. Do you plan to add support for TensorRT 8.0 in the near future or is the only solution to downgrade to Ubuntu 18.04?
NVIDIA is publishing TF1 and TF2 containers by the end of the month that comes with 20.04 and TRT8 support.
Search for "Nvidia NGC tensorflow container", the release you will be looking for will be 21.07. Released by the end of the month
NVIDIA is publishing TF1 and TF2 containers by the end of the month that comes with 20.04 and TRT8 support.
Search for "Nvidia NGC tensorflow container", the release you will be looking for will be 21.07. Released by the end of the month
Hi there, when will be the release date of TRT8 support?
Hello. The question is not quite on the topic, but I'll ask anyway. I'm trying to build a package with TensorFlow 2.9.1, and when I tried to build it with TwnsorRT 8.4 update 1 support, I got an error with the meaning that this version is not supported. Please tell me when the TF-RT bundle will support the TensorRT-8.4 version
Hello. I have the same problem as above. I'm trying to build a package with TensorFlow 2.9.1, and when I tried to build it with TensorRT 8.2 update 1 support, I got an error with the meaning that this version is not supported. Please tell me when the TF-RT bundle will support the TensorRT-8.2 version. RuntimeError: Tensorflow has not been built with TensorRT support.
I'm trying to build a package with TensorFlow 2.9.1, and when I tried to build it with TensorRT 8.2 update 1 support, I got an error with the meaning that this version is not supported. Please tell me when the TF-RT bundle will support the TensorRT-8.2 version.
My experience of building tf from source shows that using trt 8.2 u1 is quite viable. Can you show the configuration and build logs of tf?
My configuration：ubuntu 20.04, python 3.8, tensorflow 2.9.1, tensorRT 8.2.5.1, cuda 11.6 Build logs:
2022-09-06T01:55:49.4545036Z def _check_trt_version_compatibility():
2022-09-06T01:55:49.4545652Z """Check compatibility of TensorRT version.
2022-09-06T01:55:49.4546168Z
2022-09-06T01:55:49.4546616Z Raises:
2022-09-06T01:55:49.4547177Z RuntimeError: if the TensorRT library version is incompatible.
2022-09-06T01:55:49.4548022Z """
2022-09-06T01:55:49.4548510Z
2022-09-06T01:55:49.4549029Z if not _pywrap_py_utils.is_tensorrt_enabled():
2022-09-06T01:55:49.4549595Z logging.error(
2022-09-06T01:55:49.4550338Z "Tensorflow needs to be built with TensorRT support enabled to allow "
2022-09-06T01:55:49.4552311Z "TF-TRT to operate.")
2022-09-06T01:55:49.4553040Z
2022-09-06T01:55:49.4553660Z > raise RuntimeError("Tensorflow has not been built with TensorRT support.")
2022-09-06T01:55:49.4554398Z E RuntimeError: Tensorflow has not been built with TensorRT support.
May I ask what is your viable configuration please?
My configuration：ubuntu 20.04, python 3.8, tensorflow 2.9.1, tensorRT 8.2.5.1, cuda 11.6 Build logs:
2022-09-06T01:55:49.4545036Z def _check_trt_version_compatibility(): 2022-09-06T01:55:49.4545652Z """Check compatibility of TensorRT version. 2022-09-06T01:55:49.4546168Z 2022-09-06T01:55:49.4546616Z Raises: 2022-09-06T01:55:49.4547177Z RuntimeError: if the TensorRT library version is incompatible. 2022-09-06T01:55:49.4548022Z """ 2022-09-06T01:55:49.4548510Z 2022-09-06T01:55:49.4549029Z if not _pywrap_py_utils.is_tensorrt_enabled(): 2022-09-06T01:55:49.4549595Z logging.error( 2022-09-06T01:55:49.4550338Z "Tensorflow needs to be built with TensorRT support enabled to allow " 2022-09-06T01:55:49.4552311Z "TF-TRT to operate.") 2022-09-06T01:55:49.4553040Z 2022-09-06T01:55:49.4553660Z > raise RuntimeError("Tensorflow has not been built with TensorRT support.") 2022-09-06T01:55:49.4554398Z E RuntimeError: Tensorflow has not been built with TensorRT support.
May I ask what is your viable configuration please?
My configuration: Debian 11.4, Python 3.9 cuda 11.4 TRT 8.2.5.1, Bazel-5.0.0. The build was completed today. If you want to reproduce my build on Ubuntu, I recommend installing all components using the package manager (leave the Python version 3.8). Show the "./configure" stage