Decoder start token is actually 0

README.md

@@ -441,16 +441,16 @@ from transformers import T5ForConditionalGeneration, T5Tokenizer, GenerationConf
 model = T5ForConditionalGeneration.from_pretrained('jbochi/madlad400-3b-mt')
 tokenizer = T5Tokenizer.from_pretrained('jbochi/madlad400-3b-mt')
 
-text = "<2es> how do you say torch in portuguese?"
+text = "<2pt> I love pizza!"
 input_ids = tokenizer(text, return_tensors="pt").input_ids
 outputs = model.generate(
     input_ids=input_ids,
     generation_config=GenerationConfig(
-        decoder_start_token_id=2,
+        decoder_start_token_id=0,
 ))
 
 tokenizer.decode(outputs[0], skip_special_tokens=True)
-# como se dice antorcha en portugués?
+# Eu adoro pizza!
 ```
 
 Colab to generate these files is [here](https://colab.research.google.com/drive/1rZ2NRyl2zwmg0sQ2Wi-uZZF48iVYulTC#scrollTo=pVODoE6gA9sw).