[Model] support new model ovis2.5 (#23084)

Signed-off-by: myselvess <244285088@qq.com>
Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn>
Co-authored-by: Isotr0py <2037008807@qq.com>
Co-authored-by: Isotr0py <mozf@mail2.sysu.edu.cn>

tests/models/multimodal/generation/vlm_utils/model_utils.py

@@ -10,6 +10,7 @@ from typing import Optional, Union
 
 import numpy as np
 import numpy.typing as npt
+import PIL.Image
 import pytest
 import regex as re
 import torch
@@ -810,6 +811,63 @@ def ovis_patch_hf_runner(hf_model: HfRunner) -> HfRunner:
     return hf_model
 
 
+def ovis2_5_patch_hf_runner(hf_model: HfRunner) -> HfRunner:
+    """Patches and returns an instance of the HfRunner to use for Ovis2."""
+    hf_model.model.get_output_embeddings = lambda: \
+        hf_model.model.llm.get_output_embeddings()
+
+    def processor(*args, text="", images=None, videos=None, **kwargs):
+        if images is None:
+            images = []
+        else:
+            images = [images] if isinstance(images, Image) else images
+        if videos is None:
+            videos = []
+        else:
+            videos = [videos] if isinstance(videos, np.ndarray) else videos
+            videos = [[PIL.Image.fromarray(frame) for frame in vid]
+                      for vid in videos]
+
+        prompt_start_and_end = {
+            "qwen2": ("<|im_start|>user\n", "<|im_end|>\n"),
+            "llama":
+            ("<|start_header_id|>user<|end_header_id|>\n\n", "<|eot_id|>"),
+            "gemma2": ("<start_of_turn>user\n", "<end_of_turn>\n"),
+        }
+        for start, end in prompt_start_and_end.values():
+            if start in text and end in text:
+                text = text.split(start)[1].split(end)[0]
+                break
+
+        images_message = [{"type": "image", "image": img} for img in images]
+        videos_message = [{"type": "video", "video": vid} for vid in videos]
+
+        messages = [{
+            "role":
+            "user",
+            "content": [
+                *images_message,
+                *videos_message,
+                {
+                    "type": "text",
+                    "text": text
+                },
+            ],
+        }]
+
+        input_ids, pixel_values, grid_thws = hf_model.model.preprocess_inputs(
+            messages=messages, enable_thinking=True)
+        inputs = {
+            "inputs": input_ids,
+            "pixel_values": pixel_values,
+            "grid_thws": grid_thws,
+        }
+        return BatchFeature(data=inputs, tensor_type="pt")
+
+    hf_model.processor = processor
+    return hf_model
+
+
 def qwen2_5_omni_patch_hf_runner(hf_model: HfRunner) -> HfRunner:
     """Patches and returns an instance of the HfRunner for Qwen2.5-Omni."""
     thinker = hf_model.model.thinker