Fixed the MultilabelTrainer document, which would cause a potential bug when executing the code originally documented.

[Mohan-Zhang-u]

if train with the MultilabelTrainer documented in the original document

from torch import nn
from transformers import Trainer

class MultilabelTrainer(Trainer):
    def compute_loss(self, model, inputs, return_outputs=False):
        labels = inputs.pop("labels")
        outputs = model(**inputs)
        logits = outputs.logits
        loss_fct = nn.BCEWithLogitsLoss()
        loss = loss_fct(logits.view(-1, self.model.config.num_labels),
                        labels.float().view(-1, self.model.config.num_labels))
        return (loss, outputs) if return_outputs else loss

a bug that looks like this would appear:
File "~/anaconda3/lib/python3.8/site-packages/transformers/trainer.py", in train self._maybe_log_save_evaluate(tr_loss, model, trial, epoch) File "~/anaconda3/lib/python3.8/site-packages/transformers/trainer.py", in _maybe_log_save_evaluate metrics = self.evaluate() File "~/anaconda3/lib/python3.8/site-packages/transformers/trainer.py", in evaluate output = self.prediction_loop( File "~/anaconda3/lib/python3.8/site-packages/transformers/trainer.py", in prediction_loop loss, logits, labels = self.prediction_step(model, inputs, prediction_loss_only, ignore_keys=ignore_keys) File "~/anaconda3/lib/python3.8/site-packages/transformers/trainer.py", in prediction_step labels = nested_detach(tuple(inputs.get(name) for name in self.label_names)) File "~/anaconda3/lib/python3.8/site-packages/transformers/trainer_pt_utils.py", in nested_detach return type(tensors)(nested_detach(t) for t in tensors) File "~/anaconda3/lib/python3.8/site-packages/transformers/trainer_pt_utils.py", in <genexpr> return type(tensors)(nested_detach(t) for t in tensors) File "~/anaconda3/lib/python3.8/site-packages/transformers/trainer_pt_utils.py", in nested_detach return tensors.detach() AttributeError: 'NoneType' object has no attribute 'detach'

change the original code to below would effectively avoid this bug.

from torch import nn
from transformers import Trainer

class MultilabelTrainer(Trainer):
    def compute_loss(self, model, inputs, return_outputs=False):
        labels = inputs.get("labels")
        outputs = model(**inputs)
        logits = outputs.get('logits')
        loss_fct = nn.BCEWithLogitsLoss()
        loss = loss_fct(logits.view(-1, self.model.config.num_labels),
                        labels.float().view(-1, self.model.config.num_labels))
        return (loss, outputs) if return_outputs else loss

[sgugger]

Thanks for fixing!