Came across this today, and seems like it would be a little improvement.
if --use_auth_token is set but no token is available (not logged in to huggingface-cli) the program will aborts with
huggingface_hub.utils._headers.LocalTokenNotFoundError: Token is required (`token=True`), but no token found.
Even if the model specified is not gated. So should this arg be ignored and the model loaded anyway? Essentially making it okay to always include this arg. However this somehow makes this arg redundant.
can gladly open a PR to try and implement this
or is this actual behaviour that could be addressed in transformers itself?
Came across this today, and seems like it would be a little improvement.
if
--use_auth_token
is set but no token is available (not logged in to huggingface-cli) the program will aborts withEven if the model specified is not gated. So should this arg be ignored and the model loaded anyway? Essentially making it okay to always include this arg. However this somehow makes this arg redundant.
can gladly open a PR to try and implement this
or is this actual behaviour that could be addressed in transformers itself?