[Model][VLM] Add Kimi-VL model support (#16387)

Signed-off-by: courage17340 <courage17340@163.com>

examples/offline_inference/vision_language.py

@@ -364,6 +364,29 @@ def run_internvl(questions: list[str], modality: str) -> ModelRequestData:
     )
 
 
+# Kimi-VL
+def run_kimi_vl(questions: list[str], modality: str) -> ModelRequestData:
+    assert modality == "image"
+
+    prompts = [
+        "<|im_user|>user<|im_middle|><|media_start|>image<|media_content|>"
+        f"<|media_pad|><|media_end|>{question}<|im_end|>"
+        "<|im_assistant|>assistant<|im_middle|>" for question in questions
+    ]
+
+    engine_args = EngineArgs(
+        model="moonshotai/Kimi-VL-A3B-Instruct",
+        max_model_len=4096,
+        disable_mm_preprocessor_cache=args.disable_mm_preprocessor_cache,
+        trust_remote_code=True,
+    )
+
+    return ModelRequestData(
+        engine_args=engine_args,
+        prompts=prompts,
+    )
+
+
 # LLaVA-1.5
 def run_llava(questions: list[str], modality: str) -> ModelRequestData:
     assert modality == "image"
@@ -966,6 +989,7 @@ model_example_map = {
     "h2ovl_chat": run_h2ovl,
     "idefics3": run_idefics3,
     "internvl_chat": run_internvl,
+    "kimi_vl": run_kimi_vl,
     "llava": run_llava,
     "llava-next": run_llava_next,
     "llava-next-video": run_llava_next_video,

examples/offline_inference/vision_language_multi_image.py

@@ -326,6 +326,45 @@ def load_llama4(question: str, image_urls: list[str]) -> ModelRequestData:
     )
 
 
+def load_kimi_vl(question: str, image_urls: list[str]) -> ModelRequestData:
+    model_name = "moonshotai/Kimi-VL-A3B-Instruct"
+
+    engine_args = EngineArgs(
+        model=model_name,
+        max_model_len=4096,
+        max_num_seqs=4,
+        tensor_parallel_size=1,
+        limit_mm_per_prompt={"image": len(image_urls)},
+        trust_remote_code=True,
+    )
+
+    placeholders = [{"type": "image", "image": url} for url in image_urls]
+    messages = [{
+        "role":
+        "user",
+        "content": [
+            *placeholders,
+            {
+                "type": "text",
+                "text": question
+            },
+        ],
+    }]
+
+    processor = AutoProcessor.from_pretrained(model_name,
+                                              trust_remote_code=True)
+
+    prompt = processor.apply_chat_template(messages,
+                                           tokenize=False,
+                                           add_generation_prompt=True)
+
+    return ModelRequestData(
+        engine_args=engine_args,
+        prompt=prompt,
+        image_data=[fetch_image(url) for url in image_urls],
+    )
+
+
 def load_mistral3(question: str, image_urls: list[str]) -> ModelRequestData:
     model_name = "mistralai/Mistral-Small-3.1-24B-Instruct-2503"
 
@@ -640,6 +679,7 @@ model_example_map = {
     "h2ovl_chat": load_h2ovl,
     "idefics3": load_idefics3,
     "internvl_chat": load_internvl,
+    "kimi_vl": load_kimi_vl,
     "llama4": load_llama4,
     "mistral3": load_mistral3,
     "mllama": load_mllama,