feat(converse_transformation.py): fix type for bedrock cache usage block

litellm/llms/bedrock/chat/converse_transformation.py

@@ -612,8 +612,8 @@ class AmazonConverseConfig(BaseConfig):
         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"]
+        if "cacheWriteInputTokens" in usage:
+            cache_creation_input_tokens = usage["cacheWriteInputTokens"]
             input_tokens += cache_creation_input_tokens
 
         prompt_tokens_details = PromptTokensDetailsWrapper(

litellm/types/llms/bedrock.py

@@ -111,8 +111,8 @@ class ConverseTokenUsageBlock(TypedDict):
     totalTokens: int
     cacheReadInputTokenCount: int
     cacheReadInputTokens: int
-    cacheCreationInputTokenCount: int
-    cacheCreationInputTokens: int
+    cacheWriteInputTokenCount: int
+    cacheWriteInputTokens: int
 
 
 class ConverseResponseBlock(TypedDict):

tests/litellm/llms/bedrock/chat/test_converse_transformation.py

@@ -18,19 +18,27 @@ def test_transform_usage():
     usage = ConverseTokenUsageBlock(
         **{
             "cacheReadInputTokenCount": 0,
-            "cacheReadInputTokens": 10,
-            "cacheCreationInputTokenCount": 0,
-            "cacheCreationInputTokens": 0,
-            "inputTokens": 12,
-            "outputTokens": 56,
-            "totalTokens": 78,
+            "cacheReadInputTokens": 0,
+            "cacheWriteInputTokenCount": 1789,
+            "cacheWriteInputTokens": 1789,
+            "inputTokens": 3,
+            "outputTokens": 401,
+            "totalTokens": 2193,
         }
     )
     config = AmazonConverseConfig()
     openai_usage = config._transform_usage(usage)
-    assert openai_usage.prompt_tokens == 22
-    assert openai_usage.completion_tokens == 56
-    assert openai_usage.total_tokens == 78
-    assert openai_usage.prompt_tokens_details.cached_tokens == 10
-    assert openai_usage._cache_creation_input_tokens == 0
-    assert openai_usage._cache_read_input_tokens == 10
+    assert (
+        openai_usage.prompt_tokens
+        == usage["inputTokens"]
+        + usage["cacheWriteInputTokens"]
+        + usage["cacheReadInputTokens"]
+    )
+    assert openai_usage.completion_tokens == usage["outputTokens"]
+    assert openai_usage.total_tokens == usage["totalTokens"]
+    assert (
+        openai_usage.prompt_tokens_details.cached_tokens
+        == usage["cacheReadInputTokens"]
+    )
+    assert openai_usage._cache_creation_input_tokens == usage["cacheWriteInputTokens"]
+    assert openai_usage._cache_read_input_tokens == usage["cacheReadInputTokens"]