How do I transform a sequence classification model into a question answering model?

🤗Transformers
1

Hey everyone!
I have two task: first training a binary classifier and second is a QA task.
My classification model looks like this

class XLMRobertaBaseClassifier(nn.Module):
    def __init__(self):
        super(XLMRobertaBaseClassifier, self).__init__()
        self.base_model = AutoModel.from_pretrained('sismetanin/xlm_roberta_base-ru-sentiment-rusentiment')
        self.Linear = nn.Linear(768, 2)
        
    def forward(self, input_ids, token_type_ids, attention_mask):
        outputs = self.base_model(input_ids, token_type_ids=token_type_ids, attention_mask=attention_mask).pooler_output
        outputs = self.Linear(outputs)

        return outputs

and it outputs class probabilities. My question is: how I turn it to a QA model?