Generate - update cookie cutters to not initialize cache with training and...

Generate - update cookie cutters to not initialize cache with training and gradient checkpointing (#21759)

templates/adding_a_new_model/cookiecutter-template-{{cookiecutter.modelname}}/modeling_{{cookiecutter.lowercase_modelname}}.py

@@ -525,11 +525,17 @@ class {{cookiecutter.camelcase_modelname}}Encoder(nn.Module):
         output_hidden_states=False,
         return_dict=True,
     ):
+        if self.gradient_checkpointing and self.training and use_cache:
+            logger.warning(
+                "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
+            )
+            use_cache = False
+
         all_hidden_states = () if output_hidden_states else None
         all_self_attentions = () if output_attentions else None
         all_cross_attentions = () if output_attentions and self.config.add_cross_attention else None
-
         next_decoder_cache = () if use_cache else None
+
         for i, layer_module in enumerate(self.layer):
             if output_hidden_states:
                 all_hidden_states = all_hidden_states + (hidden_states,)
@@ -538,13 +544,6 @@ class {{cookiecutter.camelcase_modelname}}Encoder(nn.Module):
             past_key_value = past_key_values[i] if past_key_values is not None else None
 
             if self.gradient_checkpointing and self.training:
-
-                if use_cache:
-                    logger.warning(
-                        "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
-                    )
-                    use_cache = False
-
                 def create_custom_forward(module):
                     def custom_forward(*inputs):
                         return module(*inputs, past_key_value, output_attentions)
@@ -2525,6 +2524,10 @@ class {{cookiecutter.camelcase_modelname}}Decoder({{cookiecutter.camelcase_model
         hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)
 
         # decoder layers
+        if self.gradient_checkpointing and self.training and use_cache:
+            logger.warning("`use_cache = True` is incompatible with gradient checkpointing`. Setting `use_cache = False`...")
+            use_cache = False
+
         all_hidden_states = () if output_hidden_states else None
         all_self_attns = () if output_attentions else None
         all_cross_attentions = () if (output_attentions and encoder_hidden_states is not None) else None
@@ -2547,11 +2550,6 @@ class {{cookiecutter.camelcase_modelname}}Decoder({{cookiecutter.camelcase_model
             past_key_value = past_key_values[idx] if past_key_values is not None else None
 
             if self.gradient_checkpointing and self.training:
-
-                if use_cache:
-                    logger.warning("`use_cache = True` is incompatible with gradient checkpointing`. Setting `use_cache = False`...")
-                    use_cache = False
-
                 def create_custom_forward(module):
                     def custom_forward(*inputs):
                         # None for past_key_value