Open joey0922 opened 1 month ago
I found that text_encoder's embedding shape is 65024 while the vocab size of tokenizer is 64796. Is it not necessary to make this two value be equal? If so, how can I initialize embeddings of added new special tokens?
I found that text_encoder's embedding shape is 65024 while the vocab size of tokenizer is 64796. Is it not necessary to make this two value be equal? If so, how can I initialize embeddings of added new special tokens?