Conversation pipeline fixes (#26795)

* Adjust length limits and allow naked conversation list inputs

* Adjust length limits and allow naked conversation list inputs

* Maybe use a slightly more reasonable limit than 1024

* Skip tests for old models that never supported this anyway

* Cleanup input docstrings

* More docstring cleanup + skip failing TF test

* Make fixup

tests/models/t5/test_modeling_t5.py

@@ -891,6 +891,10 @@ class T5ModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixin,
     def test_disk_offload(self):
         pass
 
+    @unittest.skip("Does not support conversations.")
+    def test_pipeline_conversational(self):
+        pass
+
 
 class T5EncoderOnlyModelTester:
     def __init__(

tests/models/t5/test_modeling_tf_t5.py

@@ -314,6 +314,10 @@ class TFT5ModelTest(TFModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
     def test_keras_save_load(self):
         pass
 
+    @unittest.skip("Does not support conversations.")
+    def test_pipeline_conversational(self):
+        pass
+
 
 class TFT5EncoderOnlyModelTester:
     def __init__(
@@ -607,6 +611,10 @@ class TFT5GenerationIntegrationTests(unittest.TestCase):
         expected_output_string = ["Ich liebe es so sehr!", "die Transformatoren sind wirklich erstaunlich"]
         self.assertListEqual(expected_output_string, output_strings)
 
+    @unittest.skip("Does not support conversations.")
+    def test_pipeline_conversational(self):
+        pass
+
 
 @require_tf
 @require_sentencepiece