Workaround for `pip3 install torch` fails on MacOS 12 on Intel CPU with Python 3.11 about dalai HOT 3 OPEN

Unfortunately, this did not work for me in M1.

When you first run nxp dalai llama, it clones the llama.cpp repo to your home directory.

You need to remove the llama.cpp folder for the virtualenv solution to work.

from dalai.

Unfortunately, this did not work for me in M1.

@mrafaeldie12 Not sure how you mean by not work. I was able to get past at least the installation by leveraging pipenv:

% pipenv --python 3.10
% pipenv run npx dalai llama

This will force the Dalai constructor to "inherit" from the virtualenv initiated by pipenv, effectively also bypassing the need for the following commented by @alexthehurst:

Thanks for the tip! On macOS, this didn't work for me until I patched cli.js to call new Dalai('~/dalai-venv-py310').install(... instead of calling the constructor with no args. Dalai looks for Python and Pip at this path.

I'm leaving this to download and will confirm if the inference actually runs on M1 later when I get back

Edit (2023-Mar-14):

Confirmed that the inference works (but quite slow and unreliable) on an M1 machine

from dalai.

Thanks for the tip! On macOS, this didn't work for me until I patched cli.js to call new Dalai('~/dalai-venv-py310').install(... instead of calling the constructor with no args. Dalai looks for Python and Pip at this path.

from dalai.