ValueError：预期输入batch_size (2) 与目标batch_size (4) 匹配

Question

这是我正在做的文本分类任务的代码。问题似乎就出在这里。这是一个多类问题。我有3个标签。我尝试了几件事。我将标签的格式更改为整数，并尝试研究损失函数。我不确定需要更改哪些参数。

def evaluate(model, dataloader_val):
    model.eval()
    model.train(False)
    
    loss_val_total = 0
    predictions, true_vals = [], []
    
    for batch in dataloader_val:
        
        batch = tuple(b.to(device) for b in batch)
        
        inputs = {'input_ids':      batch[0],
                  'attention_mask': batch[1],
                  'labels':         batch[2],
                 }

        with torch.no_grad():        
            outputs = model(**inputs)
            
        loss = outputs[0]
        
        logits = outputs[1]
        loss_val_total += loss.item()

        probs = torch.argmax(logits, dim = 1).detach().cpu().numpy()
        label_ids = inputs['labels'].cpu().numpy()
        predictions.append(probs)
        true_vals.append(label_ids)
        
    loss_val_avg = loss_val_total/len(dataloader_val) 
    
    predictions = np.concatenate(predictions, axis=0)
    true_vals = np.concatenate(true_vals, axis=0)
    
    ### after evaluating we resume model training
    model.train(True)
    
    return loss_val_avg, predictions, true_vals

这是我得到的错误

ValueError                                Traceback (most recent call last)
<ipython-input-55-a095c6ad8f10> in <module>
     44                      }       
     45 
---> 46             outputs = model(**inputs)
ValueError: Expected input batch_size (2) to match target batch_size (4).

Answer 1

我不确定你是否故意这样做，但将标签作为模型的输入

outputs = model(**inputs)

似乎对我来说是个问题。您应该从字典

labels

中删除

inputs

。此外，如果您的模型是

BertForSequenceClassification

，那么它不会接受标签作为输入，也不会自动计算损失函数。尝试将您的代码更改为

def evaluate(model, dataloader_val):
    model.eval()
    model.train(False)

    loss_val_total = 0
    predictions, true_vals = [], []

    for batch in dataloader_val:
    
        batch = tuple(b.to(device) for b in batch)
    
        inputs = {
                    'input_ids':      batch[0],
                    'attention_mask': batch[1]
                 }
        labels = batch[2], # your labels should be given in dataloader as one-hot encoded values
                 
        with torch.no_grad():        
            outputs = model(**inputs)
        
        loss = nn.CrossEntropyLoss()(outputs, labels)
    
        loss_val_total += loss.item()

        probs = torch.argmax(outputs, dim = 1).detach().cpu().numpy()
        label_ids = torch.argmax(labels, dim = 1).cpu().numpy()
        predictions.append(probs)
        true_vals.append(label_ids)
    
    loss_val_avg = loss_val_total/len(dataloader_val) 

    predictions = np.concatenate(predictions, axis=0)
    true_vals = np.concatenate(true_vals, axis=0)

    ### after evaluating we resume model training
    model.train(True)

return loss_val_avg, predictions, true_vals