修复bug

src/baseline/trainer.py

@@ -198,14 +198,16 @@ def load_data(self) -> List[DataLoader]:
         self._test_steps = ceil(
             self._test_size / self.batch_size)  # 训练总step num
 
-        self.check_val_every_n_steps = ceil(
-            self.check_val_every_n_epoch * self._train_steps)  # 每多少个step进行验证
+
 
         # 如果是分布式训练，则步数要除以总节点数
         self._train_steps = ceil(self._train_steps / self.ddp_nodes_num)
         self._dev_steps = ceil(self._dev_steps / self.ddp_nodes_num)
         self._test_steps = ceil(self._test_steps / self.ddp_nodes_num)
 
+        self.check_val_every_n_steps = ceil(
+            self.check_val_every_n_epoch * self._train_steps)  # 每多少个step进行验证
+
         if self.check_val_every_n_steps < 10:
             self.check_val_every_n_steps = 10
 
@@ -220,23 +222,23 @@ def load_data(self) -> List[DataLoader]:
         if self.ddp_local_rank != -1:
             # 如果使用分布式训练, 对train_ds进行DistributedSampler
             train_ds = torch.utils.data.dataloader.DataLoader(
-                train_ds, sampler=DistributedSampler(train_ds), batch_size=self.batch_size, num_workers=4)
+                train_ds, sampler=DistributedSampler(train_ds), batch_size=self.batch_size, num_workers=8)
 
             dev_ds = torch.utils.data.dataloader.DataLoader(
-                dev_ds, batch_size=self.batch_size, shuffle=False, num_workers=4)
+                dev_ds, batch_size=self.batch_size, shuffle=False, num_workers=8)
 
             if test_ds is not None:
                 test_ds = torch.utils.data.dataloader.DataLoader(
-                    test_ds, batch_size=self.batch_size, shuffle=False, num_workers=4)
+                    test_ds, batch_size=self.batch_size, shuffle=False, num_workers=8)
 
         else:
             train_ds = torch.utils.data.dataloader.DataLoader(
-                train_ds, batch_size=self.batch_size, shuffle=True, num_workers=4)
+                train_ds, batch_size=self.batch_size, shuffle=True, num_workers=8)
             dev_ds = torch.utils.data.dataloader.DataLoader(
-                dev_ds, batch_size=self.batch_size, shuffle=False, num_workers=4)
+                dev_ds, batch_size=self.batch_size, shuffle=False, num_workers=8)
             if test_ds is not None:
                 test_ds = torch.utils.data.dataloader.DataLoader(
-                    test_ds, batch_size=self.batch_size, shuffle=False, num_workers=4)
+                    test_ds, batch_size=self.batch_size, shuffle=False, num_workers=8)
 
         return [train_ds, dev_ds, test_ds]
 
@@ -276,7 +278,7 @@ def load_suite(self):
         scheduler = get_linear_schedule_with_warmup(
             optimizer=optimizer,
             num_warmup_steps=self._warmup_steps,
-            num_training_steps=self._train_steps
+            num_training_steps=len(self.train_ds)*self.epochs
         ) if self._warmup_steps != -1 else None
         return model, optimizer, scheduler