fix(internvl): pass max patch count to vllm

swift/template/templates/internvl.py

@@ -30,6 +30,7 @@ def init_env_args(self):
     def replace_tag(self, media_type: Literal['image', 'video', 'audio'], index: int,
                     inputs: StdTemplateInputs) -> List[Context]:
         if self.mode == 'vllm':
+            inputs.mm_processor_kwargs.setdefault('max_dynamic_patch', self.max_num)
             image_context = ['<image>\n']
         else:
             image_context = ['<img>', [-100], '</img>\n']

tests/general/test_internvl_template.py

@@ -0,0 +1,44 @@
+import unittest
+from types import SimpleNamespace
+
+from swift.template.templates.internvl import InternvlTemplate
+
+
+class TestInternvlTemplate(unittest.TestCase):
+
+    @staticmethod
+    def _template(mode):
+        template = object.__new__(InternvlTemplate)
+        template.mode = mode
+        template.max_num = 6
+        return template
+
+    def test_vllm_receives_max_dynamic_patch(self):
+        template = self._template('vllm')
+        inputs = SimpleNamespace(mm_processor_kwargs={})
+
+        context = template.replace_tag('image', 0, inputs)
+
+        self.assertEqual(context, ['<image>\n'])
+        self.assertEqual(inputs.mm_processor_kwargs, {'max_dynamic_patch': 6})
+
+    def test_vllm_preserves_explicit_max_dynamic_patch(self):
+        template = self._template('vllm')
+        inputs = SimpleNamespace(mm_processor_kwargs={'max_dynamic_patch': 4})
+
+        template.replace_tag('image', 0, inputs)
+
+        self.assertEqual(inputs.mm_processor_kwargs, {'max_dynamic_patch': 4})
+
+    def test_transformers_does_not_receive_vllm_processor_kwargs(self):
+        template = self._template('transformers')
+        inputs = SimpleNamespace(mm_processor_kwargs={})
+
+        context = template.replace_tag('image', 0, inputs)
+
+        self.assertEqual(context, ['<img>', [-100], '</img>\n'])
+        self.assertEqual(inputs.mm_processor_kwargs, {})
+
+
+if __name__ == '__main__':
+    unittest.main()