If you managed, what tutorial you followed to do so or do you have any advice on the subject?
If you have already configured your system with the appropriate nvidia driver for your GPU, then you don’t need to do anything extra to train a model with rasa
To check if your system can detect your GPU try
nvidia-smi on your terminal. If the GPU is detected, the command should output information about all GPUs visible to your system. These GPUs will be accessible to the Rasa process as well.
For any advanced configurations(you don’t need these to start with), refer to this page on the documentation TensorFlow Configuration
Yes @dakshvar22, nvidia-smi works for me but I have to update everything and it seems like a really painful process.
I get this warnings when I train a model:
rasa_1 | 2020-03-17 22:06:17.919147: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'libnvinfer.so.6'; dlerror: libnvinfer.so.6: cannot open shared object file: No such file or directory rasa_1 | 2020-03-17 22:06:17.919256: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'libnvinfer_plugin.so.6'; dlerror: libnvinfer_plugin.so.6: cannot open shared object file: No such file or directory rasa_1 | 2020-03-17 22:06:17.919270: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:30] Cannot dlopen some TensorRT libraries. If you would like to use Nvidia GPU with TensorRT, please make sure the missing libraries mentioned above are installed properly.
I’ve been wondering if anyone successfully solved them.
@dakshvar22 Would this be a problem if I don’t have an NVIDIA Graphics card? Should I continue to use versions before 1.8? The updated pipeline contains the LexicalSyntacticFeaturizer which does not seem to work on version 1.7.1.
I’m not getting to full training completion with the new pipeline and the version 1.8.1. Looks like DIETClassifier is taking a lot of time to run.