修改 num_workers 和文件 seq2seq.md

chapter_computer-vision/semantic-segmentation-and-dataset.md

@@ -447,7 +447,7 @@ voc_test = VOCSegDataset(False, crop_size, voc_dir)
 batch_size = 64
 train_iter = gluon.data.DataLoader(voc_train, batch_size, shuffle=True,
                                    last_batch='discard',
-                                   num_workers=d2l.get_dataloader_workers())
+                                   num_workers=0)
 for X, Y in train_iter:
     print(X.shape)
     print(Y.shape)
@@ -459,7 +459,7 @@ for X, Y in train_iter:
 batch_size = 64
 train_iter = torch.utils.data.DataLoader(voc_train, batch_size, shuffle=True,
                                     drop_last=True,
-                                    num_workers=d2l.get_dataloader_workers())
+                                    num_workers=0)
 for X, Y in train_iter:
     print(X.shape)
     print(Y.shape)
@@ -472,7 +472,7 @@ batch_size = 64
 train_iter = paddle.io.DataLoader(voc_train, batch_size=batch_size, shuffle=True,
                                   drop_last=True,
                                   return_list=True,
-                                  num_workers=d2l.get_dataloader_workers())
+                                  num_workers=0)
 for X, Y in train_iter:
     print(X.shape)
     print(Y.shape)
@@ -490,7 +490,7 @@ def load_data_voc(batch_size, crop_size):
     """加载VOC语义分割数据集"""
     voc_dir = d2l.download_extract('voc2012', os.path.join(
         'VOCdevkit', 'VOC2012'))
-    num_workers = d2l.get_dataloader_workers()
+    num_workers = 0
     train_iter = gluon.data.DataLoader(
         VOCSegDataset(True, crop_size, voc_dir), batch_size,
         shuffle=True, last_batch='discard', num_workers=num_workers)
@@ -507,7 +507,7 @@ def load_data_voc(batch_size, crop_size):
     """加载VOC语义分割数据集"""
     voc_dir = d2l.download_extract('voc2012', os.path.join(
         'VOCdevkit', 'VOC2012'))
-    num_workers = d2l.get_dataloader_workers()
+    num_workers = 0
     train_iter = torch.utils.data.DataLoader(
         VOCSegDataset(True, crop_size, voc_dir), batch_size,
         shuffle=True, drop_last=True, num_workers=num_workers)
@@ -524,7 +524,7 @@ def load_data_voc(batch_size, crop_size):
     """加载VOC语义分割数据集"""
     voc_dir = d2l.download_extract('voc2012', os.path.join(
         'VOCdevkit', 'VOC2012'))
-    num_workers = d2l.get_dataloader_workers()
+    num_workers = 0
     train_iter = paddle.io.DataLoader(
         VOCSegDataset(True, crop_size, voc_dir), batch_size=batch_size,
         shuffle=True, return_list=True, drop_last=True, num_workers=num_workers)

chapter_computer-vision/semantic-segmentation-and-dataset_origin.md

@@ -383,7 +383,7 @@ Different from in image classification or object detection, labels here are thre
 batch_size = 64
 train_iter = gluon.data.DataLoader(voc_train, batch_size, shuffle=True,
                                    last_batch='discard',
-                                   num_workers=d2l.get_dataloader_workers())
+                                   num_workers=0)
 for X, Y in train_iter:
     print(X.shape)
     print(Y.shape)
@@ -395,7 +395,7 @@ for X, Y in train_iter:
 batch_size = 64
 train_iter = torch.utils.data.DataLoader(voc_train, batch_size, shuffle=True,
                                     drop_last=True,
-                                    num_workers=d2l.get_dataloader_workers())
+                                    num_workers=0)
 for X, Y in train_iter:
     print(X.shape)
     print(Y.shape)
@@ -414,7 +414,7 @@ def load_data_voc(batch_size, crop_size):
     """Load the VOC semantic segmentation dataset."""
     voc_dir = d2l.download_extract('voc2012', os.path.join(
         'VOCdevkit', 'VOC2012'))
-    num_workers = d2l.get_dataloader_workers()
+    num_workers = 0
     train_iter = gluon.data.DataLoader(
         VOCSegDataset(True, crop_size, voc_dir), batch_size,
         shuffle=True, last_batch='discard', num_workers=num_workers)
@@ -431,7 +431,7 @@ def load_data_voc(batch_size, crop_size):
     """Load the VOC semantic segmentation dataset."""
     voc_dir = d2l.download_extract('voc2012', os.path.join(
         'VOCdevkit', 'VOC2012'))
-    num_workers = d2l.get_dataloader_workers()
+    num_workers = 0
     train_iter = torch.utils.data.DataLoader(
         VOCSegDataset(True, crop_size, voc_dir), batch_size,
         shuffle=True, drop_last=True, num_workers=num_workers)

chapter_recurrent-modern/seq2seq.md

@@ -609,17 +609,20 @@ class MaskedSoftmaxCELoss(gluon.loss.SoftmaxCELoss):
 ```{.python .input}
 #@tab pytorch
 #@save
-class MaskedSoftmaxCELoss(nn.CrossEntropyLoss):
+class MaskedSoftmaxCELoss(nn.Module):
     """带遮蔽的softmax交叉熵损失函数"""
     # pred的形状：(batch_size,num_steps,vocab_size)
     # label的形状：(batch_size,num_steps)
     # valid_len的形状：(batch_size,)
+    def __init__(self, **kwargs):
+        super(MaskedSoftmaxCELoss, self).__init__(**kwargs)
+        # 初始化 nn.CrossEntropyLoss 作为类的成员
+        # reduction='none' 确保返回的是未聚合的逐元素损失，方便遮蔽
+        self.cross_entropy = nn.CrossEntropyLoss(reduction='none')
     def forward(self, pred, label, valid_len):
         weights = torch.ones_like(label)
         weights = sequence_mask(weights, valid_len)
-        self.reduction='none'
-        unweighted_loss = super(MaskedSoftmaxCELoss, self).forward(
-            pred.permute(0, 2, 1), label)
+        unweighted_loss = self.cross_entropy(pred.permute(0, 2, 1), label)
         weighted_loss = (unweighted_loss * weights).mean(dim=1)
         return weighted_loss
 ```
@@ -1120,4 +1123,4 @@ for eng, fra in zip(engs, fras):
 
 :begin_tab:`paddle`
 [Discussions](https://discuss.d2l.ai/t/11838)
-:end_tab:
+:end_tab: