[Janus] Fix processor iterating a string prompt character by character (#50044)

* Fix JanusProcessor iterating a str text input character by character

Normalize a single string to a one-element list before adding the default
system prompt or the image-generation suffix, and add a regression test.

* Remove Janus processor test per review

* Show full assertion diffs in Janus integration tests

* Refresh Janus text expectations for synthetic images

Author: tarekziade

src/transformers/models/janus/processing_janus.py

@@ -100,6 +100,9 @@ def __call__(
         )
 
         generation_mode = output_kwargs["text_kwargs"].pop("generation_mode")
+        if isinstance(text, str):
+            text = [text]
+
         if self.use_default_system_prompt and generation_mode == "text":
             text = [f"{DEFAULT_SYSTEM_PROMPT}{sample}" for sample in text]
         elif generation_mode == "image":

tests/models/janus/test_modeling_janus.py

@@ -414,6 +414,8 @@ def test_gradient_checkpointing_enable_disable(self):
 
 
 class JanusIntegrationTest(unittest.TestCase):
+    maxDiff = None
+
     def setUp(self):
         self.model_id = "deepseek-community/Janus-Pro-1B"
 
@@ -433,7 +435,7 @@ def test_model_text_generation(self):
         inputs = processor(images=image, text=prompt, generation_mode="text", return_tensors="pt").to(model.device)
 
         output = model.generate(**inputs, max_new_tokens=20, generation_mode="text", do_sample=False)
-        EXPECTED_DECODED_TEXT = 'You are a helpful language and vision assistant. You are able to understand the visual content that the user provides, and assist the user with a variety of tasks using natural language.\n\n\nDescribe what do you see here and tell me about the history behind it?\n\nThe image depicts the constellation of Leo, which is part of the zodiac and the constellation'  # fmt: skip
+        EXPECTED_DECODED_TEXT = 'You are a helpful language and vision assistant. You are able to understand the visual content that the user provides, and assist the user with a variety of tasks using natural language.\n\n\nDescribe what do you see here and tell me about the history behind it?\n\nThis image shows a night sky filled with stars, and there are silhouettes of trees at'  # fmt: skip
         text = processor.decode(output[0], skip_special_tokens=True)
         self.assertEqual(
             text,
@@ -504,7 +506,7 @@ def test_model_text_generation_with_multi_image(self):
             model.device, torch.float16
         )
 
-        EXPECTED_TEXT_COMPLETION = ["You are a helpful language and vision assistant. You are able to understand the visual content that the user provides, and assist the user with a variety of tasks using natural language.\n\nWhat do these two images  and  have in common?\n\nThe two images you provided are of the same star constellation. The first image shows the constellation of Leo, and the second image shows the constellation of Ursa Major. Both constellations are part"]  # fmt: skip
+        EXPECTED_TEXT_COMPLETION = ["You are a helpful language and vision assistant. You are able to understand the visual content that the user provides, and assist the user with a variety of tasks using natural language.\n\nWhat do these two images  and  have in common?\n\nThe image you provided shows a night sky with stars and a constellation, while the second image shows a night sky with stars and a different constellation. Both images feature stars and a dark sky,"]  # fmt: skip
         generated_ids = model.generate(**inputs, max_new_tokens=40, do_sample=False)
         text = processor.batch_decode(generated_ids, skip_special_tokens=True)
         self.assertEqual(EXPECTED_TEXT_COMPLETION, text)