Apple M1: illegal hardware instruction (using any rasa command)

@fkoerner thanks for giving this a try.

1 Like

Hi All,

I am curious if I would use docker on mac and run rasa on debian inside the docker will it work on mac with M1?


Hi @endreb, I imagine this will work, though you won’t be using the compute capability of the chip.

Hi, Reading this thread and wondering how would folks recommend I install Rasa on an M1 with Big Sur these days? According to the Python docs, 3.8.x is not fully supported on macOS 11, but the Rasa installation states it needs between version 3.6 and 3.8 (not 3.9.1). Also there are the tensor flow issues on Apple Silicon I see mentioned above. Is there a best practise, or is it best to wait and use an Intel machine with a different OS for now?

Hi @lostinlogic, I think you’re best off using the other machine.

1 Like

Hi is there any solution or workaround to this problem at all? for example does building tensorflow 2.1.0 from source work for m1? Would really appreciate if someone can provide any help


1 Like

If someone else has a solution please speak up, but I’m afraid building from source will not work. Problem is that M1-compatible tensorflow is version 2.4 and we have not yet upgraded rasa’s compatibility. This is on the roadmap for the near future, but I cannot say when we will get to it.

@HarmanDotpy, I think @endreb’s solution above should work for you, though again, you wouldn’t get the compute capability of the chip, since you wouldn’t be using the optimised tensorflow.

Any ideas when Rasa will support Tensorflow 2.4 or more??

@osterhult, I can’t say exactly, but it is on our radar and planned. I will update this post once we support 2.4 or more


Please update in this thread once it’s available.

@fkoerner Any update on the the possibility of using rasa on M1?

No updates yet, we are still ironing out some kinks. Will update this post when it’s available! :slight_smile:

Any new info on this? @fkoerner

@Bohemus307 you can see some updates here. We believe there is a performance decrease from tensorflow 2.3 to 2.5 and are seeing significantly increased training times. We are in the process of isolating the behaviour so it can be communicated to tensorflow, but since we believe this requires fixes on their end there’s some uncertainty on when this will move forward.

edit: issue here


@fkoerner So as of now, does rasa work fine on mac M1 if I install tf 2.5 ignoring the performance?

@mohit.bansal this will not work, we haven’t released the necessary code changes.

@fkoerner hi there! Is there an estimated release date available?

We just now released 2.8.9, which uses TensorFlow 2.6. We expect that users will see some performance degradation with the upgrade, in terms of runtime and memory usage. Please refer to this blogpost for more information, including what to expect, and why we upgraded anyways.

@fkoerner Hey, does that make it compatible with Apple Silicon? Still can’t seem to get it installed on an M1 system :frowning: