model: support deepseek-ocr (#11891)

Co-authored-by: yhyang201 <47235274+yhyang201@users.noreply.github.com>
Co-authored-by: yhyang201 <yhyang201@gmail.com>
Co-authored-by: Shi Shuai <126407087+shuaills@users.noreply.github.com>
Co-authored-by: Xinyuan Tong <xinyuantong.cs@gmail.com>

test/srt/test_vision_openai_server_a.py

@@ -150,6 +150,62 @@ class TestQwen2AudioServer(AudioOpenAITestMixin):
     model = "Qwen/Qwen2-Audio-7B-Instruct"
 
 
+class TestDeepseekOCRServer(TestOpenAIMLLMServerBase):
+    model = "deepseek-ai/DeepSeek-OCR"
+    trust_remote_code = False
+
+    def verify_single_image_response_for_ocr(self, response):
+        """Verify DeepSeek-OCR grounding output with coordinates"""
+        assert response.choices[0].message.role == "assistant"
+        text = response.choices[0].message.content
+        assert isinstance(text, str)
+
+        # DeepSeek-OCR uses grounding format, outputs coordinates
+        assert "text" in text.lower(), f"OCR text: {text}, should contain 'text'"
+
+        # Verify coordinate format [[x1, y1, x2, y2]]
+        import re
+
+        coord_pattern = r"\[\[[\d\s,]+\]\]"
+        assert re.search(
+            coord_pattern, text
+        ), f"OCR text: {text}, should contain coordinate format [[x1, y1, x2, y2]]"
+
+        # Verify basic response fields
+        assert response.id
+        assert response.created
+        assert response.usage.prompt_tokens > 0
+        assert response.usage.completion_tokens > 0
+        assert response.usage.total_tokens > 0
+
+    def test_single_image_chat_completion(self):
+        client = openai.Client(api_key=self.api_key, base_url=self.base_url)
+        image_url = "https://raw.githubusercontent.com/sgl-project/sgl-test-files/refs/heads/main/images/ocr-text.png"
+
+        response = client.chat.completions.create(
+            model="default",
+            messages=[
+                {
+                    "role": "user",
+                    "content": [
+                        {
+                            "type": "image_url",
+                            "image_url": {"url": image_url},
+                        },
+                        {
+                            "type": "text",
+                            "text": "<|grounding|>Convert the document to markdown.",
+                        },
+                    ],
+                },
+            ],
+            temperature=0,
+            **(self.get_vision_request_kwargs()),
+        )
+
+        self.verify_single_image_response_for_ocr(response)
+
+
 if __name__ == "__main__":
     del (
         TestOpenAIMLLMServerBase,

test/srt/test_vision_openai_server_common.py

@@ -32,6 +32,7 @@ class TestOpenAIMLLMServerBase(CustomTestCase):
     model: str
     extra_args: list = []
     fixed_args: list = ["--trust-remote-code", "--enable-multimodal"]
+    trust_remote_code: bool = True
 
     @classmethod
     def setUpClass(cls):
@@ -42,7 +43,11 @@ class TestOpenAIMLLMServerBase(CustomTestCase):
             cls.base_url,
             timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
             api_key=cls.api_key,
-            other_args=cls.extra_args + cls.fixed_args,
+            other_args=(
+                cls.extra_args + cls.fixed_args + ["--trust-remote-code"]
+                if cls.trust_remote_code
+                else []
+            ),
         )
         cls.base_url += "/v1"