Using the model potion-multilingual-128M leads to the following error: Tokenizer JSON declared unk_token="<unk>" but it’s not in the vocab on creating a StaticModel using the from_pretrained method.

See this minimal example provoking the error:

use model2vec_rs::model::StaticModel; fn main() { let model = StaticModel::from_pretrained("minishlab/potion-multilingual-128M", None, None, None) .unwrap(); let embeddings = model.encode(&["Hello World".to_string()]); dbg!(embeddings); }

CLI also fails with the same error for that model see for example:

model2vec-rs encode "Hello world" "minishlab/potion-multilingual-128M"