Merge pull request #21611 from BerriAI/litellm_perf_skip_throwaway_usage

perf: skip throwaway Usage() construction in ModelResponse.__init__

litellm/llms/bedrock/chat/invoke_transformations/amazon_qwen2_transformation.py

@@ -11,6 +11,7 @@ from typing import Any, List, Optional
 
 import httpx
 
+from litellm.types.utils import Usage
 from litellm.llms.bedrock.chat.invoke_transformations.amazon_qwen3_transformation import (
     AmazonQwen3Config,
 )
@@ -79,10 +80,11 @@ class AmazonQwen2Config(AmazonQwen3Config):
             # Set usage information if available in response
             if "usage" in response_data:
                 usage_data = response_data["usage"]
-                if hasattr(model_response, 'usage'):
-                    model_response.usage.prompt_tokens = usage_data.get("prompt_tokens", 0)
-                    model_response.usage.completion_tokens = usage_data.get("completion_tokens", 0)
-                    model_response.usage.total_tokens = usage_data.get("total_tokens", 0)
+                model_response.usage = Usage(
+                    prompt_tokens=usage_data.get("prompt_tokens", 0),
+                    completion_tokens=usage_data.get("completion_tokens", 0),
+                    total_tokens=usage_data.get("total_tokens", 0),
+                )
             
             return model_response
             

litellm/llms/bedrock/chat/invoke_transformations/amazon_qwen3_transformation.py

@@ -10,6 +10,7 @@ from typing import Any, List, Optional
 
 import httpx
 
+from litellm.types.utils import Usage
 from litellm.llms.base_llm.chat.transformation import BaseConfig
 from litellm.llms.bedrock.chat.invoke_transformations.base_invoke_transformation import (
     AmazonInvokeConfig,
@@ -201,10 +202,11 @@ class AmazonQwen3Config(AmazonInvokeConfig, BaseConfig):
             # Set usage information if available in response
             if "usage" in response_data:
                 usage_data = response_data["usage"]
-                if hasattr(model_response, 'usage'):
-                    model_response.usage.prompt_tokens = usage_data.get("prompt_tokens", 0)
-                    model_response.usage.completion_tokens = usage_data.get("completion_tokens", 0)
-                    model_response.usage.total_tokens = usage_data.get("total_tokens", 0)
+                model_response.usage = Usage(
+                    prompt_tokens=usage_data.get("prompt_tokens", 0),
+                    completion_tokens=usage_data.get("completion_tokens", 0),
+                    total_tokens=usage_data.get("total_tokens", 0),
+                )
             
             return model_response
             

litellm/types/utils.py

@@ -1826,7 +1826,7 @@ class ModelResponse(ModelResponseBase):
             else:
                 usage = usage
         elif stream is None or stream is False:
-            usage = Usage()
+            usage = None  # avoid constructing throwaway Usage; set by convert_to_model_response_object
         if hidden_params:
             self._hidden_params = hidden_params
 

tests/llm_translation/test_llm_response_utils/test_convert_dict_to_chat_completion.py

@@ -1453,3 +1453,47 @@ def test_convert_to_model_response_object_falsy_id_preserves_auto_generated(fals
     )
     assert result.id == original_id
     assert result.id.startswith("chatcmpl-")
+
+
+def test_convert_to_model_response_object_default_usage_overwritten():
+    """
+    Regression test: convert_to_model_response_object must properly set Usage
+    on a ModelResponse that only has the default Usage from ModelResponse.__init__()
+    (i.e. no extra litellm.Usage() set via setattr beforehand).
+
+    This validates the optimization of removing the redundant
+    `setattr(model_response, "usage", litellm.Usage())` in completion().
+    """
+    mr = ModelResponse()
+    # usage is not set by default (optimization: avoid constructing throwaway Usage)
+    assert not hasattr(mr, "usage")
+
+    response_object = {
+        "id": "chatcmpl-usage-test",
+        "choices": [
+            {
+                "index": 0,
+                "message": {"role": "assistant", "content": "Hello"},
+                "finish_reason": "stop",
+            }
+        ],
+        "usage": {
+            "prompt_tokens": 15,
+            "completion_tokens": 7,
+            "total_tokens": 22,
+        },
+        "model": "gpt-4o",
+    }
+
+    result = convert_to_model_response_object(
+        model_response_object=mr,
+        response_object=response_object,
+        stream=False,
+        start_time=datetime.now(),
+        end_time=datetime.now(),
+    )
+
+    assert isinstance(result, ModelResponse)
+    assert result.usage.prompt_tokens == 15
+    assert result.usage.completion_tokens == 7
+    assert result.usage.total_tokens == 22