No module named 'tensorflow.python.compiler.tensorrt'
maurera opened this issue · 2 comments
Platform: Tensorflow 2.0. Python 3.7. Testing on local windows machine with no GPU before deploying to AWS EC2 machine with GPU.
Apologies if this is a very simple question.
I'm trying to use TensorRT for the first time in order to speed up model predict time. The nvidia documentation: 2.2.4. TF-TRT 2.0 Workflow With A SavedModel says to start by running:
import tensorflow as tf
from tensorflow.python.compiler.tensorrt import trt_convert as trt
When I do this, I get the error: ModuleNotFoundError: No module named 'tensorflow.python.compiler.tensorrt'. Is tensorrt still included with tensorflow? Do I need to install it separately?
Issue solved by running on GPU machine. tensorflow.python.compiler.tensorrt appears to be included in tensorflow-gpu, but not (standard cpu-only) tensorflow
Also worth mentioning https://forums.developer.nvidia.com/t/error-converting-tensorflow-model-to-tensorrt-on-windows-10/83892 its not supported on windows 10