I am new to transformers and trying to build a question answering chatbot. I use google-t5/t5-large model and it outputs “not_duplicate” for any input question.
`

    model = AutoModelForSeq2SeqLM.from_pretrained('google-t5/t5-large').to(device)
    tokenizer = AutoTokenizer.from_pretrained('google-t5/t5-large')
    def generate_answer(question):
        input_text = f"Answer the question:
                      {question}
                      Answer:"""
        inputs = tokenizer(input_text, return_tensors="pt").input_ids.to(device)
    
        outputs = model.generate(inputs, generation_config=GenerationConfig(max_new_tokens=200, 
    num_beams=1))
        print(outputs[0])
        answer = tokenizer.decode(outputs[0], skip_special_tokens=True)
        return answer
    
    test_question = "Who is Mark Twain?"
    print(generate_answer(test_question))

and get this output

tensor([    0,    59,   834,    26,   413, 26221,     1], device='cuda:0')
not_duplicate

For other kinds of inputs, it generates at least a bad answer, but not “not_duplicate”.
Please help me out.