Mismatch: modelzoo_urls.yaml vs. commit folder name in huggingface download

[nobias]

Hi! I tried creating a project based on the "topviewmouse" model from the model zoo by calling create_pretrained_project() exactly as in the Google Collab demo notebook. It worked for me last week, but this week I got a "file not found error" when DLC tries to unzip the .tgz file downloaded from Huggingface. I didn't update my code since last week.

In the current modelzoo_urls.yaml, the commit ID for the topviewmouse model is a7d7df40c3307a3c7a0ceeb2593d46a783235b28, whereas the file created here in download_hugginface_model() seems to be placed in a subfolder that contains the commit ID 670ba95203d75b9533d6595d5cc05ce5f13d5163, hence the FileNotFoundError below.

I would guess that something changed on Hugginface (a new commit?), but was not updated in modelzoo_urls.yaml. I was able to fix the problem by replacing the commit ID starting with a7d7 with the one starting with 670b in modelzoo_urls.yaml.

Full error traceback:

---------------------------------------------------------------------------
FileNotFoundError                         Traceback (most recent call last)
Cell In[12], line 1
----> 1 config_path, train_config_path = deeplabcut.create_pretrained_project(
      2     project_name,
      3     your_name,
      4     [video_path],
      5     videotype=videotype,
      6     model=model2use,
      7     analyzevideo=True,
      8     createlabeledvideo=False,
      9     copy_videos=True, #must leave copy_videos=True
     10 )
File ~/mambaforge/envs/dlc/lib/python3.10/site-packages/deeplabcut/create_project/modelzoo.py:261, in create_pretrained_project(project, experimenter, videos, model, working_directory, copy_videos, videotype, analyzevideo, filtered, createlabeledvideo, trainFraction)
    259 # Download the weights and put then in appropriate directory
    260 print("Downloading weights...")
--> 261 download_huggingface_model(model, train_dir)
    263 pose_cfg = deeplabcut.auxiliaryfunctions.read_plainconfig(path_train_config)
    264 pose_cfg["dataset_type"] = "imgaug"
File ~/mambaforge/envs/dlc/lib/python3.10/site-packages/dlclibrary/dlcmodelzoo/modelzoo_download.py:88, in download_huggingface_model(modelname, target_dir, remove_hf_folder)
     80 hf_path = os.path.join(
     81     hf_folder,
     82     "snapshots",
     83     str(neturls[modelname + "_commit"]),
     84     targzfn,
     85 )
     87 filename = os.path.join(target_dir, hf_path)
---> 88 with tarfile.open(filename, mode="r:gz") as tar:
     89     for member in tar:
     90         if not member.isdir():
File ~/mambaforge/envs/dlc/lib/python3.10/tarfile.py:1817, in TarFile.open(cls, name, mode, fileobj, bufsize, **kwargs)
   1815     else:
   1816         raise CompressionError("unknown compression type %r" % comptype)
-> 1817     return func(name, filemode, fileobj, **kwargs)
   1819 elif "|" in mode:
   1820     filemode, comptype = mode.split("|", 1)
File ~/mambaforge/envs/dlc/lib/python3.10/tarfile.py:1863, in TarFile.gzopen(cls, name, mode, fileobj, compresslevel, **kwargs)
   1860     raise CompressionError("gzip module is not available") from None
   1862 try:
-> 1863     fileobj = GzipFile(name, mode + "b", compresslevel, fileobj)
   1864 except OSError as e:
   1865     if fileobj is not None and mode == 'r':
File ~/mambaforge/envs/dlc/lib/python3.10/gzip.py:174, in GzipFile.__init__(self, filename, mode, compresslevel, fileobj, mtime)
    172     mode += 'b'
    173 if fileobj is None:
--> 174     fileobj = self.myfileobj = builtins.open(filename, mode or 'rb')
    175 if filename is None:
    176     filename = getattr(fileobj, 'name', '')
FileNotFoundError: [Errno 2] No such file or directory: '/data0/HL/dlc-workdir/hl3-fail-H-2023-07-31/dlc-models/iteration-0/hl3-failJul31-trainset95shuffle1/train/models--mwmathis--DeepLabCutModelZoo-SuperAnimal-TopViewMouse/snapshots/a7d7df40c3307a3c7a0ceeb2593d46a783235b28/DLC_ma_supertopview5k_resnet_50_iteration-0_shuffle-1.tar.gz'

[MMathisLab]

thanks @nobias for alerting us! will fix asap!