In_Process mode for TGI transformers, edits

Author: Bryannah Hernandez

src/sagemaker/serve/builder/model_builder.py

@@ -857,7 +857,7 @@ def build(  # pylint: disable=R0911
 
     def _build_validations(self):
         """Validations needed for model server overrides, or auto-detection or fallback"""
-        if self.mode == Mode.IN_PROCESS:
+        if self.mode == Mode.IN_PROCESS and self.model_server is not ModelServer.MMS:
             raise ValueError("IN_PROCESS mode is not supported yet!")
 
         if self.inference_spec and self.model:

src/sagemaker/serve/builder/transformers_builder.py

@@ -161,7 +161,7 @@ def _get_hf_metadata_create_model(self) -> Type[Model]:
                 vpc_config=self.vpc_config,
             )
 
-        if self.mode == Mode.LOCAL_CONTAINER:
+        if self.mode == Mode.LOCAL_CONTAINER or self.mode == Mode.IN_PROCESS:
             self.image_uri = pysdk_model.serving_image_uri(
                 self.sagemaker_session.boto_region_name, "local"
             )
@@ -226,6 +226,23 @@ def _transformers_model_builder_deploy_wrapper(self, *args, **kwargs) -> Type[Pr
                 jumpstart=False,
             )
             return predictor
+        
+        if self.mode == Mode.IN_PROCESS:
+            timeout = kwargs.get("model_data_download_timeout")
+
+            predictor = TransformersLocalModePredictor(
+                self.modes[str(Mode.IN_PROCESS)], serializer, deserializer
+            )
+
+            self.modes[str(Mode.IN_PROCESS)].create_server(
+                self.image_uri,
+                timeout if timeout else DEFAULT_TIMEOUT,
+                None,
+                predictor,
+                self.pysdk_model.env,
+                jumpstart=False,
+            )
+            return predictor
 
         if "mode" in kwargs:
             del kwargs["mode"]
@@ -276,7 +293,7 @@ def _build_transformers_env(self):
 
         self.pysdk_model = self._create_transformers_model()
 
-        if self.mode == Mode.LOCAL_CONTAINER:
+        if self.mode == Mode.LOCAL_CONTAINER or self.mode == Mode.IN_PROCESS:
             self._prepare_for_mode()
 
         return self.pysdk_model

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

@@ -56,7 +56,7 @@ def _invoke_multi_model_server_serving(self, request: object, content_type: str,
         """Placeholder docstring"""
         try:
             response = requests.post(
-                f"http://{get_docker_host()}:8080/invocations",
+                f"http://{0.0.0.0}:8080/invocations",
                 data=request,
                 headers={"Content-Type": content_type, "Accept": accept},
                 timeout=600,