Regression with 0.8.0 and Windows

[davidmezzetti]

Hello, hope all is well!

I ran into the following error with TxtAI's GitHub build on Windows:

  File "D:\a\txtai\txtai\test\python\testvectors\testdense\testm2v.py", line 24, in setUpClass
    cls.model = VectorsFactory.create({"path": "minishlab/potion-base-8M"}, None)
  File "C:\hostedtoolcache\windows\Python\3.10.11\x64\lib\site-packages\txtai\vectors\dense\factory.py", line 52, in create
    return Model2Vec(config, scoring, models)
  File "C:\hostedtoolcache\windows\Python\3.10.11\x64\lib\site-packages\txtai\vectors\dense\m2v.py", line 57, in __init__
    super().__init__(config, scoring, models)
  File "C:\hostedtoolcache\windows\Python\3.10.11\x64\lib\site-packages\txtai\vectors\base.py", line 45, in __init__
    self.model = self.load(config.get("path"))
  File "C:\hostedtoolcache\windows\Python\3.10.11\x64\lib\site-packages\txtai\vectors\base.py", line 101, in load
    model = self.loadmodel(path)
  File "C:\hostedtoolcache\windows\Python\3.10.11\x64\lib\site-packages\txtai\vectors\dense\m2v.py", line 60, in loadmodel
    return StaticModel.from_pretrained(path)
  File "C:\hostedtoolcache\windows\Python\3.10.11\x64\lib\site-packages\model2vec\model.py", line 211, in from_pretrained
    return _loading_helper(
  File "C:\hostedtoolcache\windows\Python\3.10.11\x64\lib\site-packages\model2vec\model.py", line 543, in _loading_helper
    embeddings, tokenizer, config, metadata, weights, mapping = load_pretrained(
  File "C:\hostedtoolcache\windows\Python\3.10.11\x64\lib\site-packages\model2vec\persistence\persistence.py", line 104, in load_pretrained
    folder = _resolve_folder(folder_or_repo_path=folder_or_repo_path, token=token, force_download=force_download)
  File "C:\hostedtoolcache\windows\Python\3.10.11\x64\lib\site-packages\model2vec\persistence\persistence.py", line 147, in _resolve_folder
    folder = Path(huggingface_hub.snapshot_download(str(folder_or_repo_path), repo_type="model", token=token))
  File "C:\hostedtoolcache\windows\Python\3.10.11\x64\lib\site-packages\huggingface_hub\utils\_validators.py", line 106, in _inner_fn
    validate_repo_id(arg_value)
  File "C:\hostedtoolcache\windows\Python\3.10.11\x64\lib\site-packages\huggingface_hub\utils\_validators.py", line 160, in validate_repo_id
    raise HFValidationError(
huggingface_hub.errors.HFValidationError: Repo id must use alphanumeric chars, '-', '_' or '.'. The name cannot start or end with '-' or '.' and the maximum length is 96: 'minishlab\potion-base-8M'.

It seems like an error with this line or the logic here: https://github.com/MinishLab/model2vec/pull/305/changes#diff-a2534fe222d6835afb1915976f974133214d04c80e9769f49e24c89c06afb971R101 with it switching the forward slash to a backslash.

[stephantul]

Hmm yes I see. We're using pathlib but it's actually not a real path, but an identifier. Sorry about that! I'll fix it.

[davidmezzetti]

Sounds good, I appreciate it!

[stephantul]

@davidmezzetti is it possible to test the change I made? I don't have a Windows machine handy. For some reason we deleted our Windows tests, so I'll have to reconfigure those, but this will maybe provide a quick fix.

[davidmezzetti]

Is it ready to run? I noticed the windows tests passed but the linux tests failed on that PR?

[stephantul]

@davidmezzetti should be! The tests fail because of an unrelated dependency thing, I'm checking that out now.

[davidmezzetti]

Ok. I'll try and let you know.

[stephantul]

@davidmezzetti managed to simulate the bug using the Windows runner. See the chain of commits on #314. I added a new test, then reverted the fix, and this made the test fail. Then re-adding the fix also fixed the test. Would be good to get your confirmation though!

[davidmezzetti]

I also just tested and it worked https://github.com/neuml/txtai/actions/runs/23645473151/job/68876386433

I appreciate the quick fix.

[stephantul]

Alright, thanks so much for checking. We'll release this later once we've got time.

[stephantul]

@davidmezzetti We released the latest version, 0.8.1. Let me know if you have further issues, we're happy to help 🫡