Add back serialization for automatic speech recognition

Author: samruds

src/sagemaker/base_serializers.py

@@ -397,6 +397,8 @@ def serialize(self, data):
                 raise ValueError(f"Could not open/read file: {data}. {e}")
         if isinstance(data, bytes):
             return data
+        if isinstance(data, dict) and "data" in data:
+            return self.serialize(data["data"])
 
         raise ValueError(f"Object of type {type(data)} is not Data serializable.")
 

src/sagemaker/serve/builder/schema_builder.py

@@ -164,6 +164,11 @@ def _get_serializer(self, obj):
             return StringSerializer()
         if _is_jsonable(obj):
             return JSONSerializerWrapper()
+        if isinstance(obj, dict) and "content_type" in obj:
+            try:
+                return DataSerializer(content_type=obj["content_type"])
+            except ValueError as e:
+                logger.error(e)
 
         raise ValueError(
             (

tests/integ/sagemaker/serve/test_schema_builder.py

@@ -153,6 +153,7 @@ def test_model_builder_happy_path_with_task_provided_local_schema_mode(
             "ml.m5.xlarge",
         ),
         ("deepset/roberta-base-squad2", "question-answering", "ml.m5.xlarge"),
+        ("openai/whisper-large-v3", "automatic-speech-recognition", "ml.m5.xlarge")
     ],
 )
 def test_model_builder_happy_path_with_task_provided_remote_schema_mode(