IndexError: Pytorch-lightning CompositionalMetric require tensor.item() if dim=0 whether I did so

[YuyaWake]

Bug description

I am encounting the error IndexError: invalid index of a 0-dim tensor. Use tensor.item() in Python or tensor.item<T>() in C++ to convert a 0-dim tensor to a number Output is truncated.

What version are you seeing the problem on?

v2.1

How to reproduce the bug

Dataset and Model is here:


##### Dataset #####
class WrimeDataset(Dataset):
    def __init__(self, dataframe, tokenizer, max_length):
        self.tokenizer = tokenizer
        self.data = dataframe
        self.max_length = max_length

    def __len__(self):
        return len(self.data)

    def __getitem__(self, idx):
        text = self.data.iloc[idx]['input_text']
        row = self.data.iloc[idx]
        labels = row['label'] 

        # 入力テキストの前処理
        inputs = self.tokenizer(
            "感情分析：" + text,
            max_length=self.max_length,
            padding='max_length',
            truncation=True,
            return_tensors="pt"
        )

        input_ids = inputs['input_ids'].squeeze().to(dtype=torch.long)
        attention_mask = inputs['attention_mask'].squeeze().to(dtype=torch.long)

        labels = torch.tensor(labels, dtype=torch.long)
        # shape: torch.Size([8])
        
        padding_sign = torch.full((self.max_length,), -100, dtype=torch.long)
        actual_length = min(len(inputs), self.max_length)
        padding_sign[:actual_length] = input_ids[:actual_length]
        print(f'input_ids shape: {input_ids.shape}')
        print(f'attention_mask shape: {attention_mask.shape}')
        print(f'labels shape: {labels.shape}')
        return input_ids, attention_mask, labels


##### model #####
class T5Model(pl.LightningModule):
    def __init__(self, num_classes):
        super(T5Model, self).__init__()
        model_name = "sonoisa/t5-base-japanese"
        config = T5Config.from_pretrained(model_name, output_hidden_states=True)
        self.t5_model = T5ForConditionalGeneration.from_pretrained(model_name, config=config)
        self.tokenizer = T5Tokenizer.from_pretrained(model_name)
        self.acc = Accuracy(task="multiclass", num_classes=num_classes)
        self.f1 = F1Score(task="multiclass", num_classes=num_classes)
        self.qwk = CohenKappa(task="multiclass", num_classes=num_classes, weights='quadratic')
        self.class_acc = Accuracy(task='binary')
        self.class_f1 = F1Score(task='binary')
        self.classifier = nn.Linear(config.d_model, num_classes)
        self.sigmoid = nn.Sigmoid()
        self.num_classes = num_classes

        for param in self.t5_model.parameters():
            param.requires_grad = True

        for param in self.classifier.parameters():
            param.requires_grad = True
        

    def forward(self, input_ids, attention_mask=None, labels=None):
        labels = torch.tensor(labels, dtype=torch.long)
        output = self.t5_model(input_ids=input_ids, attention_mask=attention_mask, labels=labels)
        loss = output.loss
        print(f'loss shape: {loss.shape}')
        decoder_hidden_states = output.decoder_hidden_states
        last_decoder_state = decoder_hidden_states[-1][:, -1, :]
        print(f'last_hidden_state: {last_decoder_state.shape}')
        logits = self.classifier(last_decoder_state)
        print(f'logits shape: {logits.shape}')
        preds = self.sigmoid(logits)
        print(f'preds shape: {preds.shape}')
        return loss, preds, logits
    
    def training_step(self, batch, batch_idx):
        input_ids, attention_mask, labels = batch
        print('==training==')
        print(f'input_ids shape: {input_ids.shape}')
        print(f'attention_mask shape: {attention_mask.shape}')
        print(f'labels shape: {labels.shape}')
        loss, preds, logits = self.forward(input_ids=input_ids, attention_mask=attention_mask, labels=labels)
        self.log('train_loss', loss.item())
        return loss.item()
    
    def validation_step(self, batch, batch_idx):
        input_ids, attention_mask, labels = batch
        print('==validation==')
        print(f'input_ids shape: {input_ids.shape}')
        print(f'attention_mask shape: {attention_mask.shape}')
        print(f'labels shape: {labels.shape}')
        loss, preds, logits = self.forward(input_ids=input_ids, attention_mask=attention_mask, labels=labels)
        
        labels = labels.bool()
        self.acc.update(preds, labels)
        self.f1.update(preds, labels)
        self.qwk.update(preds, labels)
        
        self.log('val_loss', loss.item())

        return {'loss': loss.item(), 'preds': preds, 'labels': labels}
        
    def on_validation_epoch_end(self):
        # エポックの終わりにメトリクスの平均を計算し、ログに記録
        self.log('val_acc', self.acc.compute(), on_epoch=True, prog_bar=True)
        self.log('val_f1', self.f1.compute(), on_epoch=True, prog_bar=True)
        self.log('val_qwk', self.qwk.compute(), on_epoch=True, prog_bar=True)

        for i in range(self.num_classes):
            self.log(f'val_class_{i}_acc', self.class_acc[i].compute(), on_epoch=True, prog_bar=True)
            self.log(f'val_class_{i}_f1', self.class_f1[i].compute(), on_epoch=True, prog_bar=True)
            self.class_acc[i].reset()
            self.class_f1[i].reset()
            
        # メトリクスをリセット
        self.acc.reset()
        self.f1.reset()
        self.qwk.reset()

        

    def configure_optimizers(self):
        optimizer = Adafactor(
            self.parameters(),
            scale_parameter=True,
            relative_step=True,
            warmup_init=True,
            lr=None)  # Adafactorはlrを自動的に調整する
        return optimizer

### Error messages and logs

All error is here:
'''
---------------------------------------------------------------------------
IndexError                                Traceback (most recent call last)
Cell In[84], line 31
     28 trainer = make_trainer(max_epochs=max_epochs, name='test', patience=patience)
     30 try:
---> 31     trainer.fit(model, train_dataloader, val_dataloader)
     32 except KeyError as e:
     33     print(f"KeyError: {e}")

File c:\Users\foo\anaconda3\envs\p100_venv\lib\site-packages\pytorch_lightning\trainer\trainer.py:544, in Trainer.fit(self, model, train_dataloaders, val_dataloaders, datamodule, ckpt_path)
    542 self.state.status = TrainerStatus.RUNNING
    543 self.training = True
--> 544 call._call_and_handle_interrupt(
    545     self, self._fit_impl, model, train_dataloaders, val_dataloaders, datamodule, ckpt_path
    546 )

File c:\Users\foo\anaconda3\envs\p100_venv\lib\site-packages\pytorch_lightning\trainer\call.py:44, in _call_and_handle_interrupt(trainer, trainer_fn, *args, **kwargs)
     42     if trainer.strategy.launcher is not None:
     43         return trainer.strategy.launcher.launch(trainer_fn, *args, trainer=trainer, **kwargs)
---> 44     return trainer_fn(*args, **kwargs)
     46 except _TunerExitException:
     47     _call_teardown_hook(trainer)

File c:\Users\foo\anaconda3\envs\p100_venv\lib\site-packages\pytorch_lightning\trainer\trainer.py:580, in Trainer._fit_impl(self, model, train_dataloaders, val_dataloaders, datamodule, ckpt_path)
    573 assert self.state.fn is not None
...
   1075 def __getitem__(self, idx: int) -> "CompositionalMetric":
   1076     """Construct compositional metric using the get item operator."""
-> 1077     return CompositionalMetric(lambda x: x[idx], self, None)

IndexError: invalid index of a 0-dim tensor. Use `tensor.item()` in Python or `tensor.item<T>()` in C++ to convert a 0-dim tensor to a number
Output is truncated. View as a scrollable element or open in a text editor. Adjust cell output settings...

Environment

Current environment

``` Windows11 VSCode==1.88.1 PyTorch==2.2.2 pytorch-lightning==2.1.2 ```

More info

Please tell me what is involved this error if anyone knows solution.