Iterative generation using Input embeds and past_key_values (#35890)

* Iterative generation using input embeds

* ruff fix

* Added Testcase

* Updated comment

* ♻️ Refactored testcase

* Skip test for these models

* Continue generation using input embeds and cache

* Skip generate_continue_from_embeds test

* Refactor `prepare_input_for_generation` func

* Continue generation using input embeds and cache

* Modular changes fix

* Overwrite 'prepare_inputs_for_generation' function

tests/models/moshi/test_modeling_moshi.py

@@ -358,6 +358,10 @@ class MoshiDecoderTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMi
     def test_disk_offload_safetensors(self):
         pass
 
+    @unittest.skip(reason="Test becomes too complex with Moshi requiring multiple input modalities.")
+    def test_generate_continue_from_inputs_embeds(self):
+        pass
+
     @is_flaky(max_attempts=5, description="flaky on some models.")
     def test_save_load(self):
         super().test_save_load()
@@ -824,6 +828,7 @@ class MoshiTest(ModelTesterMixin, GenerationTesterMixin, unittest.TestCase):
             output_ids_generate = model.generate(
                 do_sample=False, max_new_tokens=self.max_new_tokens, remove_invalid_values=True
             )
+            print(output_ids_generate)
             self.assertIsNotNone(output_ids_generate)
 
     @unittest.skip(reason="The audio encoder has no gradients.")
@@ -919,6 +924,10 @@ class MoshiTest(ModelTesterMixin, GenerationTesterMixin, unittest.TestCase):
     def test_disk_offload_safetensors(self):
         pass
 
+    @unittest.skip(reason="Test becomes too complex with Moshi requiring multiple modalities")
+    def test_generate_continue_from_inputs_embeds(self):
+        pass
+
     @is_flaky(max_attempts=5, description="flaky on some models.")
     def test_save_load(self):
         super().test_save_load()