ValueError: Non-consecutive added token '<extra_id_99>' found when loading 'CogVideoXPipeline'

[daeunni]

Hi, I found the below error when I tried to load CogVideoXPipeline as follows:
pipe = CogVideoXPipeline.from_pretrained("THUDM/CogVideoX-2b").to("cuda")

--> ValueError: Non-consecutive added token '<extra_id_99>' found. Should have index 32100 but has index 32000 in saved vocabulary.

How can I address it? My diffusers version is v0.30.0. Thanks!

[tolgacangoz]

Hi @daeunni. Could you try to upgrade your transformers?
Also, if you want to use CogVideoX's latest improvements install diffusers from the source.