fix: NotRequired for file_size TypedDict field; defensive seek(0) before file_data

litellm/proxy/openai_files_endpoints/files_endpoints.py

@@ -446,6 +446,10 @@ async def create_file(  # noqa: PLR0915
         # Pass file.file (the underlying SpooledTemporaryFile / BytesIO) rather than
         # reading all bytes into memory — the IO object is fully compatible with
         # FileTypes and lets downstream providers stream the upload.
+        # Ensure pointer is at 0 before any code reads from file.file, so that
+        # subsequent seek-based operations (file_size, JSONL sniff, upload) start
+        # from the beginning of the file regardless of prior reads.
+        file.file.seek(0)
         file_data = (file.filename, file.file, file.content_type)
 
         ## check if model is a loadbalanced model

litellm/types/utils.py

@@ -41,7 +41,7 @@ from pydantic import (
     field_validator,
     model_validator,
 )
-from typing_extensions import Required, TypedDict
+from typing_extensions import NotRequired, Required, TypedDict
 
 from litellm._uuid import uuid
 from litellm.types.llms.base import (
@@ -3494,7 +3494,7 @@ class ExtractedFileData(TypedDict):
     content: Union[bytes, "IO[bytes]"]
     content_type: Optional[str]
     headers: Mapping[str, str]
-    file_size: Optional[int]
+    file_size: NotRequired[Optional[int]]
 
 
 class SpecialEnums(Enum):