Fix ConversationalPipeline tests (huggingface#26217)

Add BlenderbotSmall templates and correct handling for conversation.past_user_inputs

src/transformers/models/blenderbot_small/tokenization_blenderbot_small.py

@@ -236,3 +236,18 @@ def save_vocabulary(self, save_directory: str, filename_prefix: Optional[str] =
                 index += 1
 
         return vocab_file, merge_file
+
+    @property
+    # Copied from transformers.models.blenderbot.tokenization_blenderbot.BlenderbotTokenizer.default_chat_template
+    def default_chat_template(self):
+        """
+        A very simple chat template that just adds whitespace between messages.
+        """
+        return (
+            "{% for message in messages %}"
+            "{% if message['role'] == 'user' %}{{ ' ' }}{% endif %}"
+            "{{ message['content'] }}"
+            "{% if not loop.last %}{{ '  ' }}{% endif %}"
+            "{% endfor %}"
+            "{{ eos_token }}"
+        )

src/transformers/models/blenderbot_small/tokenization_blenderbot_small_fast.py

@@ -117,3 +117,18 @@ def create_token_type_ids_from_sequences(
         if token_ids_1 is None:
             return len(cls + token_ids_0 + sep) * [0]
         return len(cls + token_ids_0 + sep + sep + token_ids_1 + sep) * [0]
+
+    @property
+    # Copied from transformers.models.blenderbot.tokenization_blenderbot.BlenderbotTokenizer.default_chat_template
+    def default_chat_template(self):
+        """
+        A very simple chat template that just adds whitespace between messages.
+        """
+        return (
+            "{% for message in messages %}"
+            "{% if message['role'] == 'user' %}{{ ' ' }}{% endif %}"
+            "{{ message['content'] }}"
+            "{% if not loop.last %}{{ '  ' }}{% endif %}"
+            "{% endfor %}"
+            "{{ eos_token }}"
+        )

tests/pipelines/test_pipelines_conversational.py

@@ -140,8 +140,8 @@ def test_integration_torch_conversation(self):
         conversation_1 = Conversation("Going to the movies tonight - any suggestions?")
         conversation_2 = Conversation("What's the last book you have read?")
         # Then
-        self.assertEqual(len(conversation_1.past_user_inputs), 0)
-        self.assertEqual(len(conversation_2.past_user_inputs), 0)
+        self.assertEqual(len(conversation_1.past_user_inputs), 1)
+        self.assertEqual(len(conversation_2.past_user_inputs), 1)
         # When
         result = conversation_agent([conversation_1, conversation_2], do_sample=False, max_length=1000)
         # Then
@@ -171,7 +171,7 @@ def test_integration_torch_conversation_truncated_history(self):
         conversation_agent = pipeline(task="conversational", min_length_for_response=24, device=DEFAULT_DEVICE_NUM)
         conversation_1 = Conversation("Going to the movies tonight - any suggestions?")
         # Then
-        self.assertEqual(len(conversation_1.past_user_inputs), 0)
+        self.assertEqual(len(conversation_1.past_user_inputs), 1)
         # When
         result = conversation_agent(conversation_1, do_sample=False, max_length=36)
         # Then
@@ -379,8 +379,8 @@ def test_integration_torch_conversation_encoder_decoder(self):
         conversation_1 = Conversation("My name is Sarah and I live in London")
         conversation_2 = Conversation("Going to the movies tonight, What movie would you recommend? ")
         # Then
-        self.assertEqual(len(conversation_1.past_user_inputs), 0)
-        self.assertEqual(len(conversation_2.past_user_inputs), 0)
+        self.assertEqual(len(conversation_1.past_user_inputs), 1)
+        self.assertEqual(len(conversation_2.past_user_inputs), 1)
         # When
         result = conversation_agent([conversation_1, conversation_2], do_sample=False, max_length=1000)
         # Then