As noted in my commits this code is now Tensorflow 2.0 Compliant. Although it is not fully converted to 2.0; it has some pending depreciations that do not have a set removal version currently. Additionally it has 2 new files evaluation.py and evaluation2.py which is to allow the model to continue to use bidirectional_dynamicrnn and optimize_loss (They are just the files constructing the two contrib modules that were depreciated). Tensorboard indicates this is 99% GPU compliant.
Note: This is not setup for eager nor any major changes from based code. It is just a hacked together conversion to utilize the native speed boosts from newer drivers.
Dependencies:
Python 3.x (Currently using Python 3.6 with this repository)
Tensorflowgpu 2.0
CUDA 10.0 NOT 10.1
CUDNN 7.3.1 (Note: I believe you can go up to 7.5 although you'd have to fresh compile tensorflow as it is built with 7.3.1 currently)
TensorRT 5.0.2.6
**Linux Nvidia Driver 410.10 -- Unsure if backwards and forwards compatible on windows.
As noted in my commits this code is now Tensorflow 2.0 Compliant. Although it is not fully converted to 2.0; it has some pending depreciations that do not have a set removal version currently. Additionally it has 2 new files evaluation.py and evaluation2.py which is to allow the model to continue to use bidirectional_dynamicrnn and optimize_loss (They are just the files constructing the two contrib modules that were depreciated). Tensorboard indicates this is 99% GPU compliant.
Note: This is not setup for eager nor any major changes from based code. It is just a hacked together conversion to utilize the native speed boosts from newer drivers.
Dependencies: Python 3.x (Currently using Python 3.6 with this repository) Tensorflowgpu 2.0 CUDA 10.0 NOT 10.1 CUDNN 7.3.1 (Note: I believe you can go up to 7.5 although you'd have to fresh compile tensorflow as it is built with 7.3.1 currently) TensorRT 5.0.2.6 **Linux Nvidia Driver 410.10 -- Unsure if backwards and forwards compatible on windows.