I want to load a huggingface model. The model I want to load has about 150K downloads so I don't think there is any problem with the model itself.
With the both loading codes below I get the same error:
from transformers import AutoModel
AutoModel.from_pretrained("laion/CLIP-convnext_large_d_320.laion2B-s29B-b131K-ft-soup")
And
from transformers import CLIPProcessor, CLIPModel
model_id = "laion/CLIP-convnext_large_d_320.laion2B-s29B-b131K-ft-soup"
processor = CLIPProcessor.from_pretrained(model_id)
model = CLIPModel.from_pretrained(model_id)
With both I get:
OSError: laion/CLIP-convnext_large_d_320.laion2B-s29B-b131K-ft-soup does not appear to have a file named preprocessor_config.json. Checkout 'https://huggingface.co/laion/CLIP-convnext_large_d_320.laion2B-s29B-b131K-ft-soup/main' for available files.
Any help to load the model would be appreciated.
It seems this model is an OpenCLIP only model right now. You can not load it the usual way.
You should first install open_clip
with pip install open_clip_torch
then use this code:
import open_clip
model, preprocess_train, preprocess_val = open_clip.create_model_and_transforms('hf-hub:laion/CLIP-convnext_large_d_320.laion2B-s29B-b131K-ft-soup')
tokenizer = open_clip.get_tokenizer('hf-hub:laion/CLIP-convnext_large_d_320.laion2B-s29B-b131K-ft-soup')