Handle azure deepseek reasoning response

litellm/litellm_core_utils/llm_response_utils/convert_dict_to_response.py

@@ -3,6 +3,7 @@
 import time
 import traceback
 import uuid
+import re
 from typing import Dict, Iterable, List, Literal, Optional, Union
 
 import litellm
@@ -415,8 +416,17 @@ def convert_to_model_response_object(  # noqa: PLR0915
                     for field in choice["message"].keys():
                         if field not in message_keys:
                             provider_specific_fields[field] = choice["message"][field]
+
+                    # Handle reasoning models that display `reasoning_content` within `content``
+                    content = choice["message"].get("content", None)
+                    reasoning_match = re.match(r"<think>(.*?)</think>(.*)", content, re.DOTALL)
+
+                    if reasoning_match:
+                        provider_specific_fields["reasoning_content"] = reasoning_match.group(1)
+                        content = reasoning_match.group(2)
+
                     message = Message(
-                        content=choice["message"].get("content", None),
+                        content=content,
                         role=choice["message"]["role"] or "assistant",
                         function_call=choice["message"].get("function_call", None),
                         tool_calls=tool_calls,

tests/llm_translation/test_azure_ai.py

@@ -13,7 +13,7 @@
 from litellm.llms.anthropic.chat import ModelResponseIterator
 import httpx
 import json
-from respx import MockRouter
+from litellm.llms.custom_httpx.http_handler import HTTPHandler
 
 load_dotenv()
 import io
@@ -184,3 +184,45 @@ def test_completion_azure_ai_command_r():
         pass
     except Exception as e:
         pytest.fail(f"Error occurred: {e}")
+
+def test_azure_deepseek_reasoning_content():
+    import json
+
+    client = HTTPHandler()
+
+    with patch.object(client, "post") as mock_post:
+        mock_response = MagicMock()
+
+        mock_response.text = json.dumps(
+            {
+                "choices": [
+                    {
+                    "finish_reason": "stop",
+                    "index": 0,
+                    "message": {
+                        "content": "<think>I am thinking here</think>\n\nThe sky is a canvas of blue",
+                        "role": "assistant",
+                    }
+                    }
+                ],
+            }
+        ) 
+
+        mock_response.status_code = 200
+        # Add required response attributes
+        mock_response.headers = {"Content-Type": "application/json"}
+        mock_response.json = lambda: json.loads(mock_response.text)
+        mock_post.return_value = mock_response
+
+
+        response = litellm.completion(
+            model='azure_ai/deepseek-r1',
+            messages=[{"role": "user", "content": "Hello, world!"}],
+            api_base="https://litellm8397336933.services.ai.azure.com/models/chat/completions",
+            api_key="my-fake-api-key",
+            client=client
+        )  
+
+        print(response)
+        assert(response.choices[0].message.reasoning_content == "I am thinking here")
+        assert(response.choices[0].message.content == "\n\nThe sky is a canvas of blue")