fairseq/criterions/label_smoothed_cross_entropy.py

Survived

Mutant 1577

--- fairseq/criterions/label_smoothed_cross_entropy.py
+++ fairseq/criterions/label_smoothed_cross_entropy.py
@@ -9,7 +9,7 @@
 from fairseq.criterions import FairseqCriterion, register_criterion
 
 
-def label_smoothed_nll_loss(lprobs, target, epsilon, ignore_index=None, reduce=True):
+def label_smoothed_nll_loss(lprobs, target, epsilon, ignore_index=None, reduce=False):
     if target.dim() == lprobs.dim() - 1:
         target = target.unsqueeze(-1)
     nll_loss = -lprobs.gather(dim=-1, index=target)

Mutant 1578

--- fairseq/criterions/label_smoothed_cross_entropy.py
+++ fairseq/criterions/label_smoothed_cross_entropy.py
@@ -29,7 +29,7 @@
     return loss, nll_loss
 
 
-@register_criterion('label_smoothed_cross_entropy')
+@register_criterion('XXlabel_smoothed_cross_entropyXX')
 class LabelSmoothedCrossEntropyCriterion(FairseqCriterion):
 
     def __init__(self, task, sentence_avg, label_smoothing):

Mutant 1579

--- fairseq/criterions/label_smoothed_cross_entropy.py
+++ fairseq/criterions/label_smoothed_cross_entropy.py
@@ -28,8 +28,6 @@
     loss = (1. - epsilon) * nll_loss + eps_i * smooth_loss
     return loss, nll_loss
 
-
-@register_criterion('label_smoothed_cross_entropy')
 class LabelSmoothedCrossEntropyCriterion(FairseqCriterion):
 
     def __init__(self, task, sentence_avg, label_smoothing):

Mutant 1580

--- fairseq/criterions/label_smoothed_cross_entropy.py
+++ fairseq/criterions/label_smoothed_cross_entropy.py
@@ -37,7 +37,6 @@
         self.sentence_avg = sentence_avg
         self.eps = label_smoothing
 
-    @staticmethod
     def add_args(parser):
         """Add criterion-specific arguments to the parser."""
         # fmt: off

Mutant 1581

--- fairseq/criterions/label_smoothed_cross_entropy.py
+++ fairseq/criterions/label_smoothed_cross_entropy.py
@@ -45,7 +45,7 @@
                             help='epsilon for label smoothing, 0 means no label smoothing')
         # fmt: on
 
-    def forward(self, model, sample, reduce=True):
+    def forward(self, model, sample, reduce=False):
         """Compute the loss for the given sample.
 
         Returns a tuple with three elements:

Mutant 1582

--- fairseq/criterions/label_smoothed_cross_entropy.py
+++ fairseq/criterions/label_smoothed_cross_entropy.py
@@ -65,7 +65,7 @@
         }
         return loss, sample_size, logging_output
 
-    def compute_loss(self, model, net_output, sample, reduce=True):
+    def compute_loss(self, model, net_output, sample, reduce=False):
         lprobs = model.get_normalized_probs(net_output, log_probs=True)
         lprobs = lprobs.view(-1, lprobs.size(-1))
         target = model.get_targets(sample, net_output).view(-1, 1)

Mutant 1583

--- fairseq/criterions/label_smoothed_cross_entropy.py
+++ fairseq/criterions/label_smoothed_cross_entropy.py
@@ -74,7 +74,6 @@
         )
         return loss, nll_loss
 
-    @staticmethod
     def reduce_metrics(logging_outputs) -> None:
         """Aggregate logging outputs from data parallel training."""
         loss_sum = sum(log.get('loss', 0) for log in logging_outputs)

Mutant 1584

--- fairseq/criterions/label_smoothed_cross_entropy.py
+++ fairseq/criterions/label_smoothed_cross_entropy.py
@@ -86,7 +86,6 @@
         metrics.log_scalar('nll_loss', nll_loss_sum / ntokens / math.log(2), ntokens, round=3)
         metrics.log_derived('ppl', lambda meters: utils.get_perplexity(meters['nll_loss'].avg))
 
-    @staticmethod
     def logging_outputs_can_be_summed() -> bool:
         """
         Whether the logging outputs returned by `forward` can be summed