feat: InferenceSpec support for MMS and testing (#4763)

* feat: InferenceSpec support for MMS and testing

* Fix formatting

* CR Fixes for InferenceSpec MMS

* remove code

* Changes to environment, avoid duplicates

* Remove loggers and add docstring updates

* Changes for unit tests in transformers build

* formatting changes

* Add secret_key to endpoint mode

* get_model, docstring, and if changes

* pre-push fixes

* integ test edits

* formatting fixes

* format changes

* updated value error

* formatting changes for value error update

---------

Co-authored-by: Bryannah Hernandez <[email protected]>

Author: bryannahm1

src/sagemaker/serve/builder/transformers_builder.py

@@ -13,8 +13,10 @@
 """Transformers build logic with model builder"""
 from __future__ import absolute_import
 import logging
+import os
 from abc import ABC, abstractmethod
 from typing import Type
+from pathlib import Path
 from packaging.version import Version
 
 from sagemaker.model import Model
@@ -26,7 +28,12 @@
 from sagemaker.huggingface import HuggingFaceModel
 from sagemaker.serve.model_server.multi_model_server.prepare import (
     _create_dir_structure,
+    prepare_for_mms,
 )
+from sagemaker.serve.detector.image_detector import (
+    auto_detect_container,
+)
+from sagemaker.serve.detector.pickler import save_pkl
 from sagemaker.serve.utils.optimize_utils import _is_optimized
 from sagemaker.serve.utils.predictors import TransformersLocalModePredictor
 from sagemaker.serve.utils.types import ModelServer
@@ -73,6 +80,8 @@ def __init__(self):
         self.pytorch_version = None
         self.instance_type = None
         self.schema_builder = None
+        self.inference_spec = None
+        self.shared_libs = None
 
     @abstractmethod
     def _prepare_for_mode(self):
@@ -110,7 +119,7 @@ def _get_hf_metadata_create_model(self) -> Type[Model]:
         """
 
         hf_model_md = get_huggingface_model_metadata(
-            self.model, self.env_vars.get("HUGGING_FACE_HUB_TOKEN")
+            self.env_vars.get("HF_MODEL_ID"), self.env_vars.get("HUGGING_FACE_HUB_TOKEN")
         )
         hf_config = image_uris.config_for_framework("huggingface").get("inference")
         config = hf_config["versions"]
@@ -245,18 +254,22 @@ def _build_transformers_env(self):
 
         _create_dir_structure(self.model_path)
         if not hasattr(self, "pysdk_model"):
-            self.env_vars.update({"HF_MODEL_ID": self.model})
+
+            if self.inference_spec is not None:
+                self.env_vars.update({"HF_MODEL_ID": self.inference_spec.get_model()})
+            else:
+                self.env_vars.update({"HF_MODEL_ID": self.model})
 
             logger.info(self.env_vars)
 
             # TODO: Move to a helper function
             if hasattr(self.env_vars, "HF_API_TOKEN"):
                 self.hf_model_config = _get_model_config_properties_from_hf(
-                    self.model, self.env_vars.get("HF_API_TOKEN")
+                    self.env_vars.get("HF_MODEL_ID"), self.env_vars.get("HF_API_TOKEN")
                 )
             else:
                 self.hf_model_config = _get_model_config_properties_from_hf(
-                    self.model, self.env_vars.get("HUGGING_FACE_HUB_TOKEN")
+                    self.env_vars.get("HF_MODEL_ID"), self.env_vars.get("HUGGING_FACE_HUB_TOKEN")
                 )
 
         self.pysdk_model = self._create_transformers_model()
@@ -292,6 +305,42 @@ def _get_supported_version(self, hf_config, hugging_face_version, base_fw):
                 versions_to_return.append(base_fw_version)
         return sorted(versions_to_return, reverse=True)[0]
 
+    def _auto_detect_container(self):
+        """Set image_uri by detecting container via model name or inference spec"""
+        # Auto detect the container image uri
+        if self.image_uri:
+            logger.info(
+                "Skipping auto detection as the image uri is provided %s",
+                self.image_uri,
+            )
+            return
+
+        if self.model:
+            logger.info(
+                "Auto detect container url for the provided model and on instance %s",
+                self.instance_type,
+            )
+            self.image_uri = auto_detect_container(
+                self.model, self.sagemaker_session.boto_region_name, self.instance_type
+            )
+
+        elif self.inference_spec:
+            # TODO: this won't work for larger image.
+            # Fail and let the customer include the image uri
+            logger.warning(
+                "model_path provided with no image_uri. Attempting to autodetect the image\
+                    by loading the model using inference_spec.load()..."
+            )
+            self.image_uri = auto_detect_container(
+                self.inference_spec.load(self.model_path),
+                self.sagemaker_session.boto_region_name,
+                self.instance_type,
+            )
+        else:
+            raise ValueError(
+                "Cannot detect and set image_uri. Please pass model or inference spec."
+            )
+
     def _build_for_transformers(self):
         """Method that triggers model build
 
@@ -300,6 +349,26 @@ def _build_for_transformers(self):
         self.secret_key = None
         self.model_server = ModelServer.MMS
 
+        if self.inference_spec:
+
+            os.makedirs(self.model_path, exist_ok=True)
+
+            code_path = Path(self.model_path).joinpath("code")
+
+            save_pkl(code_path, (self.inference_spec, self.schema_builder))
+            logger.info("PKL file saved to file: %s", code_path)
+
+            self._auto_detect_container()
+
+            self.secret_key = prepare_for_mms(
+                model_path=self.model_path,
+                shared_libs=self.shared_libs,
+                dependencies=self.dependencies,
+                session=self.sagemaker_session,
+                image_uri=self.image_uri,
+                inference_spec=self.inference_spec,
+            )
+
         self._build_transformers_env()
 
         if self.role_arn:

src/sagemaker/serve/mode/sagemaker_endpoint_mode.py

@@ -130,6 +130,7 @@ def prepare(
                 model_path=model_path,
                 sagemaker_session=sagemaker_session,
                 s3_model_data_url=s3_model_data_url,
+                secret_key=secret_key,
                 image=image,
                 should_upload_artifacts=should_upload_artifacts,
             )

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

@@ -0,0 +1,100 @@
+"""This module is for SageMaker inference.py."""
+
+from __future__ import absolute_import
+import os
+import io
+import cloudpickle
+import shutil
+import platform
+from pathlib import Path
+from functools import partial
+from sagemaker.serve.spec.inference_spec import InferenceSpec
+from sagemaker.serve.validations.check_integrity import perform_integrity_check
+import logging
+
+logger = logging.getLogger(__name__)
+
+inference_spec = None
+schema_builder = None
+SHARED_LIBS_DIR = Path(__file__).parent.parent.joinpath("shared_libs")
+SERVE_PATH = Path(__file__).parent.joinpath("serve.pkl")
+METADATA_PATH = Path(__file__).parent.joinpath("metadata.json")
+
+
+def model_fn(model_dir):
+    """Overrides default method for loading a model"""
+    shared_libs_path = Path(model_dir + "/shared_libs")
+
+    if shared_libs_path.exists():
+        # before importing, place dynamic linked libraries in shared lib path
+        shutil.copytree(shared_libs_path, "/lib", dirs_exist_ok=True)
+
+    serve_path = Path(__file__).parent.joinpath("serve.pkl")
+    with open(str(serve_path), mode="rb") as file:
+        global inference_spec, schema_builder
+        obj = cloudpickle.load(file)
+        if isinstance(obj[0], InferenceSpec):
+            inference_spec, schema_builder = obj
+
+    if inference_spec:
+        return partial(inference_spec.invoke, model=inference_spec.load(model_dir))
+
+
+def input_fn(input_data, content_type):
+    """Deserializes the bytes that were received from the model server"""
+    try:
+        if hasattr(schema_builder, "custom_input_translator"):
+            return schema_builder.custom_input_translator.deserialize(
+                io.BytesIO(input_data), content_type
+            )
+        else:
+            return schema_builder.input_deserializer.deserialize(
+                io.BytesIO(input_data), content_type[0]
+            )
+    except Exception as e:
+        logger.error("Encountered error: %s in deserialize_response." % e)
+        raise Exception("Encountered error in deserialize_request.") from e
+
+
+def predict_fn(input_data, predict_callable):
+    """Invokes the model that is taken in by model server"""
+    return predict_callable(input_data)
+
+
+def output_fn(predictions, accept_type):
+    """Prediction is serialized to bytes and sent back to the customer"""
+    try:
+        if hasattr(schema_builder, "custom_output_translator"):
+            return schema_builder.custom_output_translator.serialize(predictions, accept_type)
+        else:
+            return schema_builder.output_serializer.serialize(predictions)
+    except Exception as e:
+        logger.error("Encountered error: %s in serialize_response." % e)
+        raise Exception("Encountered error in serialize_response.") from e
+
+
+def _run_preflight_diagnostics():
+    _py_vs_parity_check()
+    _pickle_file_integrity_check()
+
+
+def _py_vs_parity_check():
+    container_py_vs = platform.python_version()
+    local_py_vs = os.getenv("LOCAL_PYTHON")
+
+    if not local_py_vs or container_py_vs.split(".")[1] != local_py_vs.split(".")[1]:
+        logger.warning(
+            f"The local python version {local_py_vs} differs from the python version "
+            f"{container_py_vs} on the container. Please align the two to avoid unexpected behavior"
+        )
+
+
+def _pickle_file_integrity_check():
+    with open(SERVE_PATH, "rb") as f:
+        buffer = f.read()
+
+    perform_integrity_check(buffer=buffer, metadata_path=METADATA_PATH)
+
+
+# on import, execute
+_run_preflight_diagnostics()

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

@@ -14,12 +14,23 @@
 
 from __future__ import absolute_import
 import logging
-from pathlib import Path
-from typing import List
 
 from sagemaker.serve.model_server.tgi.prepare import _copy_jumpstart_artifacts
 from sagemaker.serve.utils.local_hardware import _check_disk_space, _check_docker_disk_usage
 
+from pathlib import Path
+import shutil
+from typing import List
+
+from sagemaker.session import Session
+from sagemaker.serve.spec.inference_spec import InferenceSpec
+from sagemaker.serve.detector.dependency_manager import capture_dependencies
+from sagemaker.serve.validations.check_integrity import (
+    generate_secret_key,
+    compute_hash,
+)
+from sagemaker.remote_function.core.serialization import _MetaData
+
 logger = logging.getLogger(__name__)
 
 
@@ -63,3 +74,56 @@ def prepare_mms_js_resources(
     model_path, code_dir = _create_dir_structure(model_path)
 
     return _copy_jumpstart_artifacts(model_data, js_id, code_dir)
+
+
+def prepare_for_mms(
+    model_path: str,
+    shared_libs: List[str],
+    dependencies: dict,
+    session: Session,
+    image_uri: str,
+    inference_spec: InferenceSpec = None,
+) -> str:
+    """Prepares for InferenceSpec using model_path, writes inference.py, and captures dependencies to generate secret_key.
+
+    Args:to
+        model_path (str) : Argument
+        shared_libs (List[]) : Argument
+        dependencies (dict) : Argument
+        session (Session) : Argument
+        inference_spec (InferenceSpec, optional) : Argument
+            (default is None)
+    Returns:
+        ( str ) : secret_key
+    """
+    model_path = Path(model_path)
+    if not model_path.exists():
+        model_path.mkdir()
+    elif not model_path.is_dir():
+        raise Exception("model_dir is not a valid directory")
+
+    if inference_spec:
+        inference_spec.prepare(str(model_path))
+
+    code_dir = model_path.joinpath("code")
+    code_dir.mkdir(exist_ok=True)
+
+    shutil.copy2(Path(__file__).parent.joinpath("inference.py"), code_dir)
+
+    logger.info("Finished writing inference.py to code directory")
+
+    shared_libs_dir = model_path.joinpath("shared_libs")
+    shared_libs_dir.mkdir(exist_ok=True)
+    for shared_lib in shared_libs:
+        shutil.copy2(Path(shared_lib), shared_libs_dir)
+
+    capture_dependencies(dependencies=dependencies, work_dir=code_dir)
+
+    secret_key = generate_secret_key()
+    with open(str(code_dir.joinpath("serve.pkl")), "rb") as f:
+        buffer = f.read()
+    hash_value = compute_hash(buffer=buffer, secret_key=secret_key)
+    with open(str(code_dir.joinpath("metadata.json")), "wb") as metadata:
+        metadata.write(_MetaData(hash_value).to_json())
+
+    return secret_key

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

@@ -4,6 +4,7 @@
 
 import requests
 import logging
+import platform
 from pathlib import Path
 from sagemaker import Session, fw_utils
 from sagemaker.serve.utils.exceptions import LocalModelInvocationException
@@ -31,6 +32,17 @@ def _start_serving(
         env_vars: dict,
     ):
         """Placeholder docstring"""
+        env = {
+            "SAGEMAKER_SUBMIT_DIRECTORY": "/opt/ml/model/code",
+            "SAGEMAKER_PROGRAM": "inference.py",
+            "SAGEMAKER_SERVE_SECRET_KEY": secret_key,
+            "LOCAL_PYTHON": platform.python_version(),
+        }
+        if env_vars:
+            env_vars.update(env)
+        else:
+            env_vars = env
+
         self.container = client.containers.run(
             image,
             "serve",
@@ -43,7 +55,7 @@ def _start_serving(
                     "mode": "rw",
                 },
             },
-            environment=_update_env_vars(env_vars),
+            environment=env_vars,
         )
 
     def _invoke_multi_model_server_serving(self, request: object, content_type: str, accept: str):
@@ -81,6 +93,7 @@ class SageMakerMultiModelServer:
     def _upload_server_artifacts(
         self,
         model_path: str,
+        secret_key: str,
         sagemaker_session: Session,
         s3_model_data_url: str = None,
         image: str = None,
@@ -127,6 +140,16 @@ def _upload_server_artifacts(
             else None
         )
 
+        if secret_key:
+            env_vars = {
+                "SAGEMAKER_SUBMIT_DIRECTORY": "/opt/ml/model/code",
+                "SAGEMAKER_PROGRAM": "inference.py",
+                "SAGEMAKER_SERVE_SECRET_KEY": secret_key,
+                "SAGEMAKER_REGION": sagemaker_session.boto_region_name,
+                "SAGEMAKER_CONTAINER_LOG_LEVEL": "10",
+                "LOCAL_PYTHON": platform.python_version(),
+            }
+
         return model_data, _update_env_vars(env_vars)
 
 

src/sagemaker/serve/spec/inference_spec.py

@@ -30,3 +30,6 @@ def invoke(self, input_object: object, model: object):
 
     def prepare(self, *args, **kwargs):
         """Custom prepare function"""
+
+    def get_model(self):
+        """Return HuggingFace model name for inference spec"""