diff --git a/examples/offline_inference/vision_language.py b/examples/offline_inference/vision_language.py
index 1d6d819ff58a..9fd9da3b0855 100644
--- a/examples/offline_inference/vision_language.py
+++ b/examples/offline_inference/vision_language.py
@@ -1140,14 +1140,10 @@ def run_ovis2_5(questions: list[str], modality: str) -> ModelRequestData:
     elif modality == "video":
         placeholder = "<video>"
 
-    tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
-    messages = [
-        [{"role": "user", "content": f"{placeholder}\n{question}"}]
+    prompts = [
+        f"<|im_start|>user\n\n{placeholder}\n{question}<|im_end|>\n<|im_start|>assistant\n"
         for question in questions
     ]
-    prompts = tokenizer.apply_chat_template(
-        messages, tokenize=False, add_generation_prompt=True
-    )
 
     return ModelRequestData(
         engine_args=engine_args,
diff --git a/examples/offline_inference/vision_language_multi_image.py b/examples/offline_inference/vision_language_multi_image.py
index 3e711e205313..bdadad0f68b1 100644
--- a/examples/offline_inference/vision_language_multi_image.py
+++ b/examples/offline_inference/vision_language_multi_image.py
@@ -713,12 +713,7 @@ def load_ovis2_5(question: str, image_urls: list[str]) -> ModelRequestData:
     placeholders = "\n".join(
         f"Image-{i}: <image>\n" for i, _ in enumerate(image_urls, start=1)
     )
-    messages = [{"role": "user", "content": f"{placeholders}\n{question}"}]
-
-    tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
-    prompt = tokenizer.apply_chat_template(
-        messages, tokenize=False, add_generation_prompt=True
-    )
+    prompt = f"<|im_start|>user\n\n{placeholders}\n{question}<|im_end|>\n<|im_start|>assistant\n"
 
     return ModelRequestData(
         engine_args=engine_args,
diff --git a/vllm/model_executor/models/ovis.py b/vllm/model_executor/models/ovis.py
index 08ce8c5d83a6..d120c1fbb2c5 100644
--- a/vllm/model_executor/models/ovis.py
+++ b/vllm/model_executor/models/ovis.py
@@ -414,6 +414,8 @@ def get_replacement_ovis(item_idx: int):
     dummy_inputs=OvisDummyInputsBuilder,
 )
 class Ovis(nn.Module, SupportsMultiModal, SupportsPP):
+    merge_by_field_config = True
+
     @classmethod
     def get_placeholder_str(cls, modality: str, i: int) -> Optional[str]:
         if modality.startswith("image"):
diff --git a/vllm/model_executor/models/ovis2_5.py b/vllm/model_executor/models/ovis2_5.py
index 8f73f2ff8263..da4f07c69131 100644
--- a/vllm/model_executor/models/ovis2_5.py
+++ b/vllm/model_executor/models/ovis2_5.py
@@ -380,7 +380,9 @@ def _call_hf_processor(
                 self.visual_indicators_to_visual_tokens(indicator)
                 for indicator in visual_indicators
             ]
-            processed_outputs["video_indicator_tokens"] = indicator_tokens
+            processed_outputs["video_indicator_tokens"] = torch.tensor(
+                [indicator_tokens]
+            )
         if "images" in mm_data:
             visual_indicators = [
                 hf_processor.construct_visual_indicators((1, 1, 1), False)
@@ -391,7 +393,7 @@ def _call_hf_processor(
                 for indicator in visual_indicators
             ]
 
-            processed_outputs["indicator_tokens"] = indicator_tokens
+            processed_outputs["indicator_tokens"] = torch.tensor([indicator_tokens])
         return processed_outputs
 
     def _apply_hf_processor_tokens_only(
@@ -441,6 +443,8 @@ def get_replacement_ovis(item_idx, modality: str):
     dummy_inputs=Ovis2_5DummyInputsBuilder,
 )
 class Ovis2_5(nn.Module, SupportsMultiModal, SupportsPP):
+    merge_by_field_config = True
+
     def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
         super().__init__()
         config = vllm_config.model_config.hf_config
@@ -491,13 +495,13 @@ def _parse_and_validate_image_input(
 
             return OvisImagePatchInputs(
                 type="image_patches",
-                flat_data=flatten_bn(flatten_bn(pixel_values), concat=True),
+                flat_data=pixel_values,
                 patches_per_image=[
                     x.shape[0] // (self.config.vit_config.hidden_stride**2)
                     for x in flatten_bn(pixel_values)
                 ],
-                indicator_tokens=flatten_bn(flatten_bn(indicator_tokens), concat=True),
-                grids=flatten_bn(flatten_bn(grids), concat=True),
+                indicator_tokens=indicator_tokens,
+                grids=grids,
             )
 
         raise AssertionError("This line should be unreachable.")
@@ -525,13 +529,13 @@ def _parse_and_validate_video_input(
 
             return OvisVideoPatchInputs(
                 type="video_patches",
-                flat_data=flatten_bn(flatten_bn(pixel_values), concat=True),
+                flat_data=flatten_bn(pixel_values, concat=True),
                 patches_per_image=[
                     x.shape[0] // (self.config.vit_config.hidden_stride**2)
-                    for x in flatten_bn(pixel_values)
+                    for x in pixel_values
                 ],
-                indicator_tokens=flatten_bn(flatten_bn(indicator_tokens), concat=True),
-                grids=flatten_bn(flatten_bn(grids), concat=True),
+                indicator_tokens=flatten_bn(indicator_tokens, concat=True),
+                grids=flatten_bn(grids, concat=True),
             )
 
         raise AssertionError("This line should be unreachable.")