fix(bedrock): count UTF-8 bytes, not string length (#36388)

litellm/llms/bedrock/files/transformation.py

@@ -863,7 +863,13 @@ class BedrockFilesConfig(BaseAWSLLM, BaseFilesConfig):
         litellm_params["upload_url"] = api_base
         # Store file size for response transformation (S3 PUT response
         # doesn't include Content-Length, so we pass it through).
-        litellm_params["_file_content_size"] = len(file_content) if isinstance(file_content, (str, bytes)) else 0
+        # Use UTF-8 byte length, not decoded string length, so multi-byte
+        # characters are counted correctly.
+        if isinstance(file_content, bytes):
+            _file_content_bytes = len(file_content)
+        else:
+            _file_content_bytes = len(file_content.encode("utf-8"))
+        litellm_params["_file_content_size"] = _file_content_bytes
 
         # Return a dict that tells the HTTP handler exactly what to do
         return {

tests/test_litellm/llms/bedrock/files/test_bedrock_files_transformation.py

@@ -2093,6 +2093,38 @@ class TestBedrockFilesS3SignatureEncoding:
 
         assert result.bytes == 2048
 
+    def test_file_content_size_counts_utf8_bytes_not_characters(self):
+        """Multi-byte characters must count as their encoded byte length.
+
+        Regression test for Greptile review on #36388: len() on a decoded
+        string undercounts multi-byte (e.g. CJK) content.
+        """
+        import httpx
+
+        from litellm.llms.bedrock.files.transformation import BedrockFilesConfig
+
+        raw_response = httpx.Response(
+            status_code=200,
+            headers={},  # S3 PUT omits Content-Length
+            content=b"",
+            request=httpx.Request(
+                "PUT",
+                "https://s3.us-east-1.amazonaws.com/my-bucket/litellm-bedrock-files-abc.jsonl",
+            ),
+        )
+
+        # 3 CJK characters = 9 UTF-8 bytes, not 3 characters.
+        content_size = len("日本語".encode("utf-8"))
+
+        result = BedrockFilesConfig().transform_create_file_response(
+            model=None,
+            raw_response=raw_response,
+            logging_obj=MagicMock(),
+            litellm_params={"_file_content_size": content_size},
+        )
+
+        assert result.bytes == 9
+
     def test_file_content_signs_spaced_object_key_the_way_s3_does(
         self, monkeypatch: pytest.MonkeyPatch
     ) -> None: