[Model] Support Dots OCR (#24645)

Signed-off-by: Roger Wang <hey@rogerw.io>
Co-authored-by: yinz-aizip <yinz@aizip.ai>

examples/offline_inference/vision_language.py

@@ -126,6 +126,23 @@ def run_chameleon(questions: list[str], modality: str) -> ModelRequestData:
     )
 
 
+# Dots-OCR
+def run_dots_ocr(questions: list[str], modality: str) -> ModelRequestData:
+    assert modality == "image"
+
+    prompts = [f"<|img|><|imgpad|><|endofimg|>{question}" for question in questions]
+    engine_args = EngineArgs(
+        model="rednote-hilab/dots.ocr",
+        limit_mm_per_prompt={modality: 1},
+        trust_remote_code=True,
+    )
+
+    return ModelRequestData(
+        engine_args=engine_args,
+        prompts=prompts,
+    )
+
+
 def run_command_a_vision(questions: list[str], modality: str) -> ModelRequestData:
     assert modality == "image"
 
@@ -1676,6 +1693,7 @@ model_example_map = {
     "aya_vision": run_aya_vision,
     "blip-2": run_blip2,
     "chameleon": run_chameleon,
+    "dots_ocr": run_dots_ocr,
     "command_a_vision": run_command_a_vision,
     "deepseek_vl_v2": run_deepseek_vl2,
     "ernie45_vl": run_ernie45_vl,