more post_init

Author: Cyrilvallez

src/transformers/models/layoutlmv3/modeling_layoutlmv3.py

@@ -591,7 +591,7 @@ def __init__(self, config):
 
         self.encoder = LayoutLMv3Encoder(config)
 
-        self.init_weights()
+        self.post_init()
 
     def get_input_embeddings(self):
         return self.embeddings.word_embeddings
@@ -881,7 +881,7 @@ def __init__(self, config):
         else:
             self.classifier = LayoutLMv3ClassificationHead(config, pool_feature=False)
 
-        self.init_weights()
+        self.post_init()
 
     @auto_docstring
     def forward(
@@ -980,7 +980,7 @@ def __init__(self, config):
         self.layoutlmv3 = LayoutLMv3Model(config)
         self.qa_outputs = LayoutLMv3ClassificationHead(config, pool_feature=False)
 
-        self.init_weights()
+        self.post_init()
 
     @auto_docstring
     def forward(
@@ -1099,7 +1099,7 @@ def __init__(self, config):
         self.layoutlmv3 = LayoutLMv3Model(config)
         self.classifier = LayoutLMv3ClassificationHead(config, pool_feature=False)
 
-        self.init_weights()
+        self.post_init()
 
     @auto_docstring
     def forward(

src/transformers/models/udop/modeling_udop.py

@@ -1074,6 +1074,7 @@ def __init__(self, config):
 
         # get weights from encoder position bias
         self.relative_bias = self._get_relative_bias(config)
+        self.post_init()
 
     @staticmethod
     def _get_relative_bias(config: UdopConfig) -> RelativePositionBiasAggregated:

tests/trainer/test_trainer.py

@@ -447,6 +447,7 @@ def __init__(self, config):
             self.a = nn.Parameter(torch.tensor(config.a).float())
             self.b = nn.Parameter(torch.tensor(config.b).float())
             self.double_output = config.double_output
+            self.post_init()
 
         def forward(self, input_x, labels=None, **kwargs):
             y = input_x * self.a + self.b
@@ -466,6 +467,7 @@ def __init__(self, config):
             self.head = nn.Linear(config.hidden_size, 1)
             self.gradient_checkpointing = False
             self.double_output = config.double_output
+            self.post_init()
 
         def forward(self, input_x, labels=None, **kwargs):
             y = input_x.unsqueeze(0)
@@ -496,6 +498,7 @@ def __init__(self, config):
             self.a = nn.Parameter(torch.tensor(config.a).float())
             self.b = nn.Parameter(torch.tensor(config.b).float())
             self.random_torch = config.random_torch
+            self.post_init()
 
         def forward(self, input_x, labels=None, **kwargs):
             y = input_x * self.a + self.b