Add forward method to dummy models (#14419)

* Add forward method to dummy models

* Fix quality

utils/check_dummies.py

@@ -43,6 +43,30 @@ class {0}:
         requires_backends(cls, {1})
 """
 
+PT_DUMMY_PRETRAINED_CLASS = (
+    DUMMY_PRETRAINED_CLASS
+    + """
+    def forward(self, *args, **kwargs):
+        requires_backends(self, {1})
+"""
+)
+
+TF_DUMMY_PRETRAINED_CLASS = (
+    DUMMY_PRETRAINED_CLASS
+    + """
+    def call(self, *args, **kwargs):
+        requires_backends(self, {1})
+"""
+)
+
+FLAX_DUMMY_PRETRAINED_CLASS = (
+    DUMMY_PRETRAINED_CLASS
+    + """
+    def __call__(self, *args, **kwargs):
+        requires_backends(self, {1})
+"""
+)
+
 DUMMY_CLASS = """
 class {0}:
     def __init__(self, *args, **kwargs):
@@ -102,8 +126,7 @@ def read_init():
 
 def create_dummy_object(name, backend_name):
     """Create the code for the dummy object corresponding to `name`."""
-    _pretrained = [
-        "Config",
+    _models = [
         "ForCausalLM",
         "ForConditionalGeneration",
         "ForMaskedLM",
@@ -114,14 +137,24 @@ def create_dummy_object(name, backend_name):
         "ForSequenceClassification",
         "ForTokenClassification",
         "Model",
-        "Tokenizer",
-        "Processor",
     ]
+    _pretrained = ["Config", "Tokenizer", "Processor"]
     if name.isupper():
         return DUMMY_CONSTANT.format(name)
     elif name.islower():
         return DUMMY_FUNCTION.format(name, backend_name)
     else:
+        is_model = False
+        for part in _models:
+            if part in name:
+                is_model = True
+                break
+        if is_model:
+            if name.startswith("TF"):
+                return TF_DUMMY_PRETRAINED_CLASS.format(name, backend_name)
+            if name.startswith("Flax"):
+                return FLAX_DUMMY_PRETRAINED_CLASS.format(name, backend_name)
+            return PT_DUMMY_PRETRAINED_CLASS.format(name, backend_name)
         is_pretrained = False
         for part in _pretrained:
             if part in name: