Difference between vocab_size in model T5forConditionalGeneration "t5-small" and its corresponding Tokenizer "t5-small"

There appears to be a difference between model.config.vocab_size and tokenizer.vocab_size for T5ForConditionalGeneration - t5-small. Not sure where the additional 28 tokens appear from. Why is that so? Bart does not have this issue.

2 Likes