Rasa-Pro setup AVX Issue on macOS M1 - Rasa CALM - Rasa Community Forum

Rasa-Pro setup AVX Issue on macOS M1

post by tal-yifat on Nov 12, 2024

Hello Rasa Community!

I’m encountering an issue while trying to run rasa-pro on my MacBook with an M1 chip (macOS Sequoia). When I run rasa init --template tutorial, I get the following error:

The TensorFlow library was compiled to use AVX instructions, but these aren't available on your machine.
Abort trap: 6

This error message seems to be because TensorFlow’s default binary distribution was built for x86 processors and optimized with AVX support, while Apple M1 chip (and subsequent Apple Silicon chips) does not support AVX instructions.

Context:

• OS: macOS Sequoia 15.1 (M1 chip)
• rasa-pro version: 3.10.8
• rasa-sdk version: 3.10.0
• Python version: 3.10.15
• TensorFlow version: 2.14.1
• Deployment Method: Local installation following the Rasa Pro tutorial, using pip for environment, and CLI to run rasa init --template tutorial. I have also tried with rasa init and got the same results.

Steps I’m considering:

• re-installing the environment with Poetry instead of pip.
• re-installing the environment with TensorFlow for macOS with Apple silicon support: tensorflow-macos and tensorflow-metal.
• Using Rosetta 2 for x86 emulation (potential performance concerns).
• Docker container with x86 emulation (potential performance concerns).

I’d like to know if anyone has found a reliable solution to this issue based on their experience with Rasa on macOS M1. Is there an approach that minimizes compatibility issues without significant performance trade-offs?

Thank you in advance for your insights!

post by stephens on Nov 19, 2024

Sounds like the wrong tensorflow. My install steps on an M1:

uv venv
source .venv/bin/activate
uv pip install rasa-pro==3.10.10
uv pip list | grep tensorflow
safetensors                                       0.4.5
tensorboard                                       2.14.1
tensorboard-data-server                           0.7.2
tensorflow-estimator                              2.14.0
tensorflow-hub                                    0.13.0
tensorflow-io-gcs-filesystem                      0.37.1
tensorflow-macos                                  2.14.1