feat(converse_transformation.py): translate converse usage block with cache creation values to openai format

litellm/llms/bedrock/chat/converse_transformation.py

@@ -31,7 +31,7 @@ from litellm.types.llms.openai import (
     ChatCompletionUserMessage,
     OpenAIMessageContentListBlock,
 )
-from litellm.types.utils import ModelResponse, Usage
+from litellm.types.utils import ModelResponse, PromptTokensDetailsWrapper, Usage
 from litellm.utils import add_dummy_tool, has_tool_call_blocks
 
 from ..common_utils import BedrockError, BedrockModelInfo, get_bedrock_tool_name
@@ -602,6 +602,33 @@ class AmazonConverseConfig(BaseConfig):
                 thinking_blocks_list.append(_thinking_block)
         return thinking_blocks_list
 
+    def _transform_usage(self, usage: ConverseTokenUsageBlock) -> Usage:
+        input_tokens = usage["inputTokens"]
+        output_tokens = usage["outputTokens"]
+        total_tokens = usage["totalTokens"]
+        cache_creation_input_tokens: int = 0
+        cache_read_input_tokens: int = 0
+
+        if "cacheReadInputTokens" in usage:
+            cache_read_input_tokens = usage["cacheReadInputTokens"]
+            input_tokens += cache_read_input_tokens
+        if "cacheCreationInputTokens" in usage:
+            cache_creation_input_tokens = usage["cacheCreationInputTokens"]
+            input_tokens += cache_creation_input_tokens
+
+        prompt_tokens_details = PromptTokensDetailsWrapper(
+            cached_tokens=cache_read_input_tokens
+        )
+        openai_usage = Usage(
+            prompt_tokens=input_tokens,
+            completion_tokens=output_tokens,
+            total_tokens=total_tokens,
+            prompt_tokens_details=prompt_tokens_details,
+            cache_creation_input_tokens=cache_creation_input_tokens,
+            cache_read_input_tokens=cache_read_input_tokens,
+        )
+        return openai_usage
+
     def _transform_response(
         self,
         model: str,
@@ -730,9 +757,7 @@ class AmazonConverseConfig(BaseConfig):
             chat_completion_message["tool_calls"] = tools
 
         ## CALCULATING USAGE - bedrock returns usage in the headers
-        input_tokens = completion_response["usage"]["inputTokens"]
-        output_tokens = completion_response["usage"]["outputTokens"]
-        total_tokens = completion_response["usage"]["totalTokens"]
+        usage = self._transform_usage(completion_response["usage"])
 
         model_response.choices = [
             litellm.Choices(
@@ -743,11 +768,7 @@ class AmazonConverseConfig(BaseConfig):
         ]
         model_response.created = int(time.time())
         model_response.model = model
-        usage = Usage(
-            prompt_tokens=input_tokens,
-            completion_tokens=output_tokens,
-            total_tokens=total_tokens,
-        )
+
         setattr(model_response, "usage", usage)
 
         # Add "trace" from Bedrock guardrails - if user has opted in to returning it

litellm/types/llms/bedrock.py

@@ -109,6 +109,10 @@ class ConverseTokenUsageBlock(TypedDict):
     inputTokens: int
     outputTokens: int
     totalTokens: int
+    cacheReadInputTokenCount: int
+    cacheReadInputTokens: int
+    cacheCreationInputTokenCount: int
+    cacheCreationInputTokens: int
 
 
 class ConverseResponseBlock(TypedDict):
@@ -400,7 +404,9 @@ class AmazonNovaCanvasTextToImageParams(TypedDict, total=False):
     conditionImage: str
 
 
-class AmazonNovaCanvasTextToImageRequest(AmazonNovaCanvasRequestBase, TypedDict, total=False):
+class AmazonNovaCanvasTextToImageRequest(
+    AmazonNovaCanvasRequestBase, TypedDict, total=False
+):
     """
     Request for Amazon Nova Canvas Text to Image API
 

tests/llm_translation/test_bedrock_completion.py

@@ -2948,3 +2948,12 @@ async def test_bedrock_stream_thinking_content_openwebui():
     assert (
         len(response_content) > 0
     ), "There should be non-empty content after thinking tags"
+
+
+def test_bedrock_usage_block():
+    litellm._turn_on_debug()
+    response = completion(
+        model="bedrock/us.anthropic.claude-3-7-sonnet-20250219-v1:0",
+        messages=[{"role": "user", "content": "Hello who is this?"}],
+    )
+    assert response.usage.total_tokens > 0