The decoder appears to be adding an extra space at the end of the string:

tokenizer.decode(tokenizer.encode('I can feel the magic, can you?').ids) produces 'I can feel the magic, can you? ' instead of 'I can feel the magic, can you?'