MNT Move to huggingface_hub v0.10 #162
Conversation
Resolves skops-dev#155 There are two major changes accompanying this move: - Model cards have now moved to the hub libary, modelcards is deprecated - Some API changes around token => use_auth_token
|
Hmm, CI is failing on Windows with encoding errors for model cards. I wonder if that is caused by something getting lost on the move to HF hub. E.g. here, the encoding used to be indicated explicitly: https://github.com/nateraw/modelcards/blob/119c895eaf97554e82f6cd73d4fd0e9aa289c3c7/modelcards/cards.py#L82 The save and load methods of Ping @nateraw @merveenoyan |
|
What's the solution here? Should I update the encoding in Would be really nice if we had windows tests there too but guessing that's a rather large undertaking. CC @Wauplin - wdyt? |
|
We rather wait till the issue is fixed and release (in a minor match maybe) on the |
That restriction was a short term solution because we thought we could make v0.10 work with a few changes. However, that's not the case because of a bug in HF hub model cards on Windows. This requires a bugfix release before it would be fixed. In the meantime, we should allow users to install v0.10, since there is no technical reason not to use it. The only annoyance is that users will get FutureWarnings. For CI, we filter out corresponding FutureWarnings for the time being. Once the HF hub bug is fixed, we can proceed with skops-dev#162 and remove the filter.
|
Hi there 👋 Thanks for reporting this. I created a PR to fix this (huggingface/huggingface_hub#1102) and will make a patch release as suggested. As for having the CI run on a windows machine as well for (and sorry for late reply, I was off last week :) ) |
|
@BenjaminBossan @adrinjalali I just made a patch release for |
|
@Wauplin I bumped the min version of |
adrinjalali
changed the title
|
@BenjaminBossan Perfect ! |
Resolves #155
There are two major changes accompanying this move:
modelcardsis deprecatedtoken=>use_auth_tokenSome docstrings were fixed too.
Comment: Took me several tries to get the
token=>use_auth_tokentransition right, as it's hard to guess what method uses what, and the use of**kwargsby some methods is not helping.