How to use rasa without the tensorflow pipeline

I am trying to use the rasa lib to create a chatbot and am getting the following error while trying to create the project:

The TensorFlow library was compiled to use AVX instructions, but these aren't available on your machine.

So I would like to know how I can install the rasa lib without using the tensorflow pipeline.

Can anyone help?

Hi @emanuel, welcome to the forum!

This is strange. I thought by default TF is compiled without AVX. I think this might not be a rasa installation issue. Did you install TF before?

I actually used this command that I found in the documentation:

pip install rasa-x --extra-index-url https://pypi.rasa.com/simple

Ok. I actually think it is just a warning that you can safely ignore. Or do you encounter another related error?

Yes, I can’t train or run a basic assistant.

The system keeps returning the same error.

Ok, AVX as default was introduced with TF 1.6. If your CPU doesn’t support AVX, you’ll have to build TF from source before installing rasa (or afterwards using pip instll -U). This issue might help: https://github.com/tensorflow/tensorflow/issues/19584

After compiling a version of TF how do I install rasa and make it use that compiled version?

Hello, I was able to compile tensforflow from source, but when I install rasa it still uses the default version. How do I get rasa to use the compiled version of tf?

You’ll have to have it in the same virtual / conda environment as rasa. I’ve never done it though.