model server might have already done a serialization. honor that by not decoding the request again if it is not already bytes or bytestream

Author: gwang111

src/sagemaker/serve/model_server/multi_model_server/inference.py

@@ -46,18 +46,24 @@ def input_fn(input_data, content_type, context=None):
         if hasattr(schema_builder, "custom_input_translator"):
             deserialized_data = schema_builder.custom_input_translator.deserialize(
                 (
-                    io.BytesIO(input_data)
-                    if type(input_data) == bytes
-                    else io.BytesIO(input_data.encode("utf-8"))
+                    input_data
+                    if not any([
+                        isinstance(input_data, bytes),
+                        isinstance(input_data, bytearray),
+                    ])
+                    else io.BytesIO(input_data)
                 ),
                 content_type,
             )
         else:
             deserialized_data = schema_builder.input_deserializer.deserialize(
                 (
-                    io.BytesIO(input_data)
-                    if type(input_data) == bytes
-                    else io.BytesIO(input_data.encode("utf-8"))
+                    input_data
+                    if not any([
+                        isinstance(input_data, bytes),
+                        isinstance(input_data, bytearray),
+                    ])
+                    else io.BytesIO(input_data)
                 ),
                 content_type[0],
             )

src/sagemaker/serve/model_server/torchserve/inference.py

@@ -68,18 +68,24 @@ def input_fn(input_data, content_type):
         if hasattr(schema_builder, "custom_input_translator"):
             deserialized_data = schema_builder.custom_input_translator.deserialize(
                 (
-                    io.BytesIO(input_data)
-                    if type(input_data) == bytes
-                    else io.BytesIO(input_data.encode("utf-8"))
+                    input_data
+                    if not any([
+                        isinstance(input_data, bytes),
+                        isinstance(input_data, bytearray),
+                    ])
+                    else io.BytesIO(input_data)
                 ),
                 content_type,
             )
         else:
             deserialized_data = schema_builder.input_deserializer.deserialize(
                 (
-                    io.BytesIO(input_data)
-                    if type(input_data) == bytes
-                    else io.BytesIO(input_data.encode("utf-8"))
+                    input_data
+                    if not any([
+                        isinstance(input_data, bytes),
+                        isinstance(input_data, bytearray),
+                    ])
+                    else io.BytesIO(input_data)
                 ),
                 content_type[0],
             )

src/sagemaker/serve/model_server/torchserve/xgboost_inference.py

@@ -71,18 +71,24 @@ def input_fn(input_data, content_type):
         if hasattr(schema_builder, "custom_input_translator"):
             return schema_builder.custom_input_translator.deserialize(
                 (
-                    io.BytesIO(input_data)
-                    if type(input_data) == bytes
-                    else io.BytesIO(input_data.encode("utf-8"))
+                    input_data
+                    if not any([
+                        isinstance(input_data, bytes),
+                        isinstance(input_data, bytearray),
+                    ])
+                    else io.BytesIO(input_data)
                 ),
                 content_type,
             )
         else:
             return schema_builder.input_deserializer.deserialize(
                 (
-                    io.BytesIO(input_data)
-                    if type(input_data) == bytes
-                    else io.BytesIO(input_data.encode("utf-8"))
+                    input_data
+                    if not any([
+                        isinstance(input_data, bytes),
+                        isinstance(input_data, bytearray),
+                    ])
+                    else io.BytesIO(input_data)
                 ),
                 content_type[0],
             )