Fix image-text-to-text stop_sequence handling

src/transformers/pipelines/image_text_to_text.py

@@ -160,13 +160,12 @@ def _sanitize_parameters(
         if generate_kwargs is not None:
             forward_kwargs["generate_kwargs"] = generate_kwargs
         if stop_sequence is not None:
-            stop_sequence_ids = self.processor.tokenizer.encode(stop_sequence, add_special_tokens=False)
-            if len(stop_sequence_ids) > 1:
-                logger.warning_once(
-                    "Stopping on a multiple token sequence is not yet supported on transformers. The first token of"
-                    " the stop sequence will be used as the stop sequence string in the interim."
-                )
-            generate_kwargs["eos_token_id"] = stop_sequence_ids[0]
+            if generate_kwargs is None:
+                generate_kwargs = {}
+            if isinstance(stop_sequence, str):
+                stop_sequence = [stop_sequence]
+            generate_kwargs["stop_strings"] = stop_sequence
+            generate_kwargs["tokenizer"] = self.processor.tokenizer
         if generate_kwargs is not None:
             forward_kwargs["generate_kwargs"] = generate_kwargs
         if max_new_tokens is not None:

tests/pipelines/test_pipelines_image_text_to_text.py

@@ -14,6 +14,7 @@
 
 import base64
 import unittest
+from types import SimpleNamespace
 
 from transformers import MODEL_FOR_IMAGE_TEXT_TO_TEXT_MAPPING, is_vision_available
 from transformers.pipelines import ImageTextToTextPipeline, pipeline
@@ -68,6 +69,17 @@ def run_pipeline_test(self, pipe, examples):
             ],
         )
 
+    def test_stop_sequence_without_generate_kwargs(self):
+        pipe = object.__new__(ImageTextToTextPipeline)
+        tokenizer = object()
+        pipe.processor = SimpleNamespace(tokenizer=tokenizer)
+
+        _, forward_kwargs, _ = pipe._sanitize_parameters(stop_sequence=".", max_new_tokens=3)
+        self.assertEqual(
+            forward_kwargs["generate_kwargs"],
+            {"stop_strings": ["."], "tokenizer": tokenizer, "max_new_tokens": 3},
+        )
+
     @require_torch
     def test_small_model_pt_token_text_only(self):
         pipe = pipeline("image-text-to-text", model="llava-hf/llava-interleave-qwen-0.5b-hf")