downloading the model for the first time is very slow

[aviv1ron1]

For a new user, who tried to run the code for the first time, downloading the base model from hf programmatically is very slow.
its much faster to pre-download it using the hf cli (also need to make sure you are logged in to hf).
recommend to add this to the documentation, as first impression is critical and user's will just give up.

[lastras]

My impression is that using an HF token is very important - maybe because HF is throttling code being executed in colab.

[aviv1ron1]

yes, if you are not logged in, hf impose an agressive rate limit on the download

[aviv1ron1]

but even if you are logged in, the hf cli uses multi threading and downloads much faster

[lastras]

ok!!