[Generation, Gemma 3] When passing a custom generation_config, overwrite default values with the model's base generation_config (#36684)

tests/models/cohere2/test_modeling_cohere2.py

@@ -16,6 +16,7 @@
 
 import unittest
 
+import pytest
 from packaging import version
 from parameterized import parameterized
 from pytest import mark
@@ -81,6 +82,7 @@ class Cohere2ModelTest(CohereModelTest, unittest.TestCase):
         pass
 
     @parameterized.expand([("random",), ("same",)])
+    @pytest.mark.generate
     @unittest.skip("Cohere2 has HybridCache which is not compatible with assisted decoding")
     def test_assisted_decoding_matches_greedy_search(self, assistant_type):
         pass
@@ -89,6 +91,7 @@ class Cohere2ModelTest(CohereModelTest, unittest.TestCase):
     def test_prompt_lookup_decoding_matches_greedy_search(self, assistant_type):
         pass
 
+    @pytest.mark.generate
     @unittest.skip("Cohere2 has HybridCache which is not compatible with assisted decoding")
     def test_assisted_decoding_sample(self):
         pass