Allow to specify train_bs and eval_bs separately in hapi.fit() (#48032)

* Fix hAPI bug of not compatible with LayerHook

https://github.com/PaddlePaddle/Paddle/issues/47000

* Fix hAPI bug of not compatible with LayerHook

* Allow to specify train_bs and eval_bs separately in hapi.fit()

* Update model.py

* Update Model.py

* Update test_model.py

* update model.py

python/paddle/hapi/model.py

@@ -1713,7 +1713,7 @@ class Model:
                 evaluation at the end of epoch. If None, will not do evaluation.
                 An instance of paddle.io.Dataset or paddle.io.Dataloader
                 is recomended. Default: None.
-            batch_size (int, optional): The batch size of train_data and eval_data. When
+            batch_size (int|list, optional): The batch size of train_data and eval_data. When
                 train_data and eval_data are both the instance of Dataloader, this
                 parameter will be ignored. Default: 1.
             epochs (int, optional): The number of epochs to train the model. Default: 1.
@@ -1836,10 +1836,20 @@ class Model:
         """
         assert train_data is not None, "train_data must be given!"
 
+        if isinstance(batch_size, (tuple, list)) and all(
+            [isinstance(x, int) for x in batch_size]
+        ):
+            assert (
+                len(batch_size) == 2
+            ), "batch_size length error, expected train_batch_size and eval_batch_size."
+            train_batch_size, eval_batch_size = batch_size
+        elif isinstance(batch_size, int):
+            train_batch_size, eval_batch_size = batch_size, batch_size
+
         if isinstance(train_data, Dataset):
             train_sampler = DistributedBatchSampler(
                 train_data,
-                batch_size=batch_size,
+                batch_size=train_batch_size,
                 shuffle=shuffle,
                 drop_last=drop_last,
             )
@@ -1855,7 +1865,7 @@ class Model:
 
         if eval_data is not None and isinstance(eval_data, Dataset):
             eval_sampler = DistributedBatchSampler(
-                eval_data, batch_size=batch_size
+                eval_data, batch_size=eval_batch_size
             )
             eval_loader = DataLoader(
                 eval_data,

python/paddle/tests/test_model.py

@@ -312,6 +312,8 @@ class TestModel(unittest.TestCase):
             self.val_dataset, batch_size=64, num_iters=num_iters
         )
 
+        model.fit(self.train_dataset, batch_size=(64, 64), shuffle=False)
+
         train_sampler = DistributedBatchSampler(
             self.train_dataset,
             batch_size=64,