Yes @Rasa PLEASE support RASA/Tensorflow operations for the new Apple M1 chips. We can try setting up a virtual machine, but it’s far less convenient than being able to run simple operations locally.
I myself get in trouble with Rasa and Numpy version when using Apple’s TF initiative. So I was hoping that you at Rasa could make a little effort and see if you guys could make this work if it is just as simple as supporting the right version of Numpy (or some other libraries).
See [Apple M1 processor and Rasa].
Hi @osterhult, sorry for the delay, I have taken this on and am working on a branch, but among other things I had some hurdles updating to Big Sur (without which I could not install tensorflow_macos). I will update this when I have more information.
Hi @osterhult, @Chris98106 sorry for the delay again, I have been trying to get this to work but unfortunately it seems that using tensorflow-macos is not possible due to the tensorflow version the branch is based on. The tensorflow-macos version is based on r2.4rc0, which is incompatible with current rasa (module attributes have changed since the last rasa-compatible version 2.3).
We’ve looked into updating the tensorflow dependency for rasa, but cannot due to a bug on the tensorflow side. There’s an issue for the bug here. If tensorflow-macos expands support to other tensorflow versions, or the above bug is fixed, the situation may change, in which case I will be happy to give this another shot. Thank you for your patience with this!