chore: Update huggingface_hub classes used after library upgrade #7631
Conversation
vblagoje
added
the
ignore-for-release-notes
vblagoje
requested review from
dfokina and
silvanocerza
and removed request for
a team
|
One sec @silvanocerza not sure how this reno note ended up here |
|
@vblagoje no worries, it's some autoformatting add a new line at the end of the file. Feel free to leave it there. |
|
Ok @silvanocerza let's see if CI is green now |
| @@ -208,8 +208,8 @@ def _run_non_streaming(self, prompt: str, generation_kwargs: Dict[str, Any]): | |||
| meta = [ | |||
| { | |||
| "model": self._client.model, | |||
| "finish_reason": tgr.details.finish_reason, | |||
| "usage": {"completion_tokens": len(tgr.details.tokens)}, | |||
| "finish_reason": tgr.details.finish_reason if tgr.details else None, | |||
There was a problem hiding this comment.
@silvanocerza this was failing in local tests because details is now optional field. I added this guard here, in chat version and TGI equivalents. Double check that guarding logic please
|
@silvanocerza I tested HuggingFaceTGIGenerator, HuggingFaceAPIGenerator, HuggingFaceTGIChatGenerator, HuggingFaceAPIChatGenerator in both streaming and non-streaming variants - they all work now. |
|
This PR required reno note, we don't want users to somehow end up using the old version of huggingface_hub which will cause issues. |
vblagoje
removed
the
ignore-for-release-notes
Upgraded the required version of
huggingface_hubto>=0.23.0across various modules to ensure compatibility and leverage the latest features. This update includes modifications to error handling for token generation details and introduces adjustments in the chat and text generation interfaces to enhance functionality and developer experience. Users are advised to upgrade theirhuggingface_hubdependency.