Error when loading

#1
by nicolas-dufour - opened

When loading the tokenizer with :

transformers.CLIPTokenizer.from_pretrained("facebook/metaclip-h14-fullcc2.5b")

i get the following error:
ValueError: Non-consecutive added token '<|startoftext|>' found. Should have index 49408 but has index 49406 in saved vocabulary.

After upgrading transformers, tokenizers and huggingface-hub this issue went away for me.

Sign up or log in to comment