launch v2 bundles API

docs/concepts/model_bundles.md

@@ -5,18 +5,22 @@ are created by packaging a model up into a deployable format.
 
 ## Creating Model Bundles
 
-There are three methods for creating model bundles:
+There are four methods for creating model bundles:
 [`create_model_bundle_from_callable_v2`](/api/client/#launch.client.LaunchClient.create_model_bundle_from_callable_v2),
 [`create_model_bundle_from_dirs_v2`](/api/client/#launch.client.LaunchClient.create_model_bundle_from_dirs_v2),
+[`create_model_bundle_from_runnable_image_v2`](/api/client/#launch.client.LaunchClient.create_model_bundle_from_runnable_image_v2),
 and
-[`create_model_bundle_from_runnable_image_v2`](/api/client/#launch.client.LaunchClient.create_model_bundle_from_runnable_image_v2).
+[`create_model_bundle_from_triton_enhanced_runnable_image_v2`](/api/client/#launch.client.LaunchClient.create_model_bundle_from_triton_enhanced_runnable_image_v2).
+
 The first directly pickles a user-specified `load_predict_fn`, a function which
 loads the model and returns a `predict_fn`, a function which takes in a request.
 The second takes in directories containing a `load_predict_fn` and the
 module path to the `load_predict_fn`.
 The third takes a Docker image and a command that starts a process listening for
 requests at port 5005 using HTTP and exposes `POST /predict` and
-`GET /healthz` endpoints.
+`GET /readyz` endpoints.
+The fourth is a variant of the third that also starts an instance of the NVidia
+Triton framework for efficient model serving.
 
 Each of these modes of creating a model bundle is called a "Flavor".
 
@@ -43,6 +47,12 @@ Each of these modes of creating a model bundle is called a "Flavor".
     * You are comfortable with building a web server and Docker image to serve your model.
 
 
+    A `TritonEnhancedRunnableImageFlavor` (a runnable image variant) is good if:
+
+    * You want to use a `RunnableImageFlavor`
+    * You also want to use [NVidia's `tritonserver`](https://developer.nvidia.com/nvidia-triton-inference-server) to accelerate model inference
+
+
 === "Creating From Callables"
     ```py
     import os
@@ -178,36 +188,66 @@ Each of these modes of creating a model bundle is called a "Flavor".
 
     BUNDLE_PARAMS = {
         "model_bundle_name": "test-bundle",
-        "load_model_fn": my_load_model_fn,
-        "load_predict_fn": my_load_predict_fn,
         "request_schema": MyRequestSchema,
         "response_schema": MyResponseSchema,
-        "repository": "launch_rearch",
-        "tag": "12b9131c5a1489c76592cddd186962cce965f0f6-cpu",
+        "repository": "...",
+        "tag": "...",
         "command": [
-            "dumb-init",
-            "--",
-            "ddtrace-run",
-            "run-service",
-            "--config",
-            "/install/launch_rearch/config/service--user_defined_code.yaml",
-            "--concurrency",
-            "1",
-            "--http",
-            "production",
-            "--port",
-            "5005",
+            ...
         ],
         "env": {
             "TEST_KEY": "test_value",
         },
+        "readiness_initial_delay_seconds": 30,
     }
 
     client = LaunchClient(api_key=os.getenv("LAUNCH_API_KEY"))
     client.create_model_bundle_from_runnable_image_v2(**BUNDLE_PARAMS)
     ```
 
 
+=== "Creating From a Triton Enhanced Runnable Image"
+    ```py
+    import os
+    from pydantic import BaseModel
+    from launch import LaunchClient
+
+
+    class MyRequestSchema(BaseModel):
+        x: int
+        y: str
+
+    class MyResponseSchema(BaseModel):
+        __root__: int
+
+
+    BUNDLE_PARAMS = {
+        "model_bundle_name": "test-bundle",
+        "request_schema": MyRequestSchema,
+        "response_schema": MyResponseSchema,
+        "repository": "...",
+        "tag": "...",
+        "command": [
+            ...
+        ],
+        "env": {
+            "TEST_KEY": "test_value",
+        },
+        "readiness_initial_delay_seconds": 30,
+        "triton_model_repository": "...",
+        "triton_model_replicas": {"": ""},
+        "triton_num_cpu": 4.0,
+        "triton_commit_tag": "",
+        "triton_storage": "",
+        "triton_memory": "",
+        "triton_readiness_initial_delay_seconds": 300,
+    }
+
+    client = LaunchClient(api_key=os.getenv("LAUNCH_API_KEY"))
+    client.create_model_bundle_from_triton_enhanced_runnable_image_v2(**BUNDLE_PARAMS)
+    ```
+
+
 ## Configuring Model Bundles
 
 The `app_config` field of a model bundle is a dictionary that can be used to

docs/index.md

@@ -33,22 +33,20 @@ def my_load_predict_fn(model):
     return returns_model_of_x_plus_len_of_y
 
 
-def my_model(x):
-    return x * 2
-
-ENV_PARAMS = {
-    "framework_type": "pytorch",
-    "pytorch_image_tag": "1.7.1-cuda11.0-cudnn8-runtime",
-}
+def my_load_model_fn():
+    def my_model(x):
+        return x * 2
+
+    return my_model
 
 BUNDLE_PARAMS = {
     "model_bundle_name": "test-bundle",
-    "model": my_model,
     "load_predict_fn": my_load_predict_fn,
-    "env_params": ENV_PARAMS,
-    "requirements": ["pytest==7.2.1", "numpy"],  # list your requirements here
+    "load_model_fn": my_load_model_fn,
     "request_schema": MyRequestSchema,
     "response_schema": MyResponseSchema,
+    "requirements": ["pytest==7.2.1", "numpy"],  # list your requirements here
+    "pytorch_image_tag": "1.7.1-cuda11.0-cudnn8-runtime",
 }
 
 ENDPOINT_PARAMS = {
@@ -81,7 +79,7 @@ def predict_on_endpoint(request: MyRequestSchema) -> MyResponseSchema:
 
 client = LaunchClient(api_key=os.getenv("LAUNCH_API_KEY"))
 
-client.create_model_bundle(**BUNDLE_PARAMS)
+client.create_model_bundle_from_callable_v2(**BUNDLE_PARAMS)
 endpoint = client.create_model_endpoint(**ENDPOINT_PARAMS)
 
 request = MyRequestSchema(x=5, y="hello")

launch/api_client/api_client.py

@@ -416,9 +416,7 @@ def __init__(
         self.content = content
 
     def _serialize_json(
-        self,
-        in_data: typing.Union[None, int, float, str, bool, dict, list],
-        eliminate_whitespace: bool = False,
+        self, in_data: typing.Union[None, int, float, str, bool, dict, list], eliminate_whitespace: bool = False
     ) -> str:
         if eliminate_whitespace:
             return json.dumps(in_data, separators=self._json_encoder.compact_separators)
@@ -483,19 +481,7 @@ def __serialize_simple(
     def serialize(
         self,
         in_data: typing.Union[
-            Schema,
-            Decimal,
-            int,
-            float,
-            str,
-            date,
-            datetime,
-            None,
-            bool,
-            list,
-            tuple,
-            dict,
-            frozendict.frozendict,
+            Schema, Decimal, int, float, str, date, datetime, None, bool, list, tuple, dict, frozendict.frozendict
         ],
     ) -> typing.Dict[str, str]:
         if self.schema:
@@ -611,19 +597,7 @@ def get_prefix_separator_iterator(self) -> typing.Optional[PrefixSeparatorIterat
     def serialize(
         self,
         in_data: typing.Union[
-            Schema,
-            Decimal,
-            int,
-            float,
-            str,
-            date,
-            datetime,
-            None,
-            bool,
-            list,
-            tuple,
-            dict,
-            frozendict.frozendict,
+            Schema, Decimal, int, float, str, date, datetime, None, bool, list, tuple, dict, frozendict.frozendict
         ],
         prefix_separator_iterator: typing.Optional[PrefixSeparatorIterator] = None,
     ) -> typing.Dict[str, str]:
@@ -691,19 +665,7 @@ def __init__(
     def serialize(
         self,
         in_data: typing.Union[
-            Schema,
-            Decimal,
-            int,
-            float,
-            str,
-            date,
-            datetime,
-            None,
-            bool,
-            list,
-            tuple,
-            dict,
-            frozendict.frozendict,
+            Schema, Decimal, int, float, str, date, datetime, None, bool, list, tuple, dict, frozendict.frozendict
         ],
     ) -> typing.Dict[str, str]:
         if self.schema:
@@ -770,19 +732,7 @@ def __to_headers(in_data: typing.Tuple[typing.Tuple[str, str], ...]) -> HTTPHead
     def serialize(
         self,
         in_data: typing.Union[
-            Schema,
-            Decimal,
-            int,
-            float,
-            str,
-            date,
-            datetime,
-            None,
-            bool,
-            list,
-            tuple,
-            dict,
-            frozendict.frozendict,
+            Schema, Decimal, int, float, str, date, datetime, None, bool, list, tuple, dict, frozendict.frozendict
         ],
     ) -> HTTPHeaderDict:
         if self.schema:
@@ -940,9 +890,7 @@ def __deserialize_application_octet_stream(
             return response.data
 
     @staticmethod
-    def __deserialize_multipart_form_data(
-        response: urllib3.HTTPResponse,
-    ) -> typing.Dict[str, typing.Any]:
+    def __deserialize_multipart_form_data(response: urllib3.HTTPResponse) -> typing.Dict[str, typing.Any]:
         msg = email.message_from_bytes(response.data)
         return {
             part.get_param("name", header="Content-Disposition"): part.get_payload(decode=True).decode(
@@ -1295,9 +1243,7 @@ def _verify_typed_dict_inputs_oapg(
         if required_keys_with_unset_values:
             raise ApiValueError(
                 "{} contains invalid unset values for {} required keys: {}".format(
-                    cls.__name__,
-                    len(required_keys_with_unset_values),
-                    required_keys_with_unset_values,
+                    cls.__name__, len(required_keys_with_unset_values), required_keys_with_unset_values
                 )
             )
 

launch/api_client/model/batch_job_serialization_format.pyi

@@ -19,7 +19,8 @@ from datetime import date, datetime  # noqa: F401
 
 import frozendict  # noqa: F401
 import typing_extensions  # noqa: F401
-from launch_client import schemas  # noqa: F401
+
+from launch.api_client import schemas  # noqa: F401
 
 class BatchJobSerializationFormat(schemas.EnumBase, schemas.StrSchema):
     """NOTE: This class is auto generated by OpenAPI Generator.

launch/api_client/model/batch_job_status.pyi

@@ -19,7 +19,8 @@ from datetime import date, datetime  # noqa: F401
 
 import frozendict  # noqa: F401
 import typing_extensions  # noqa: F401
-from launch_client import schemas  # noqa: F401
+
+from launch.api_client import schemas  # noqa: F401
 
 class BatchJobStatus(schemas.EnumBase, schemas.StrSchema):
     """NOTE: This class is auto generated by OpenAPI Generator.

launch/api_client/model/callback_auth.pyi

@@ -19,7 +19,8 @@ from datetime import date, datetime  # noqa: F401
 
 import frozendict  # noqa: F401
 import typing_extensions  # noqa: F401
-from launch_client import schemas  # noqa: F401
+
+from launch.api_client import schemas  # noqa: F401
 
 class CallbackAuth(
     schemas.ComposedSchema,
@@ -100,5 +101,5 @@ class CallbackAuth(
             **kwargs,
         )
 
-from launch_client.model.callback_basic_auth import CallbackBasicAuth
-from launch_client.model.callbackm_tls_auth import CallbackmTLSAuth
+from launch.api_client.model.callback_basic_auth import CallbackBasicAuth
+from launch.api_client.model.callbackm_tls_auth import CallbackmTLSAuth

launch/api_client/model/callback_basic_auth.pyi

@@ -19,7 +19,8 @@ from datetime import date, datetime  # noqa: F401
 
 import frozendict  # noqa: F401
 import typing_extensions  # noqa: F401
-from launch_client import schemas  # noqa: F401
+
+from launch.api_client import schemas  # noqa: F401
 
 class CallbackBasicAuth(schemas.DictSchema):
     """NOTE: This class is auto generated by OpenAPI Generator.

launch/api_client/model/callbackm_tls_auth.pyi

@@ -19,7 +19,8 @@ from datetime import date, datetime  # noqa: F401
 
 import frozendict  # noqa: F401
 import typing_extensions  # noqa: F401
-from launch_client import schemas  # noqa: F401
+
+from launch.api_client import schemas  # noqa: F401
 
 class CallbackmTLSAuth(schemas.DictSchema):
     """NOTE: This class is auto generated by OpenAPI Generator.

launch/api_client/model/clone_model_bundle_v1_request.pyi

@@ -19,7 +19,8 @@ from datetime import date, datetime  # noqa: F401
 
 import frozendict  # noqa: F401
 import typing_extensions  # noqa: F401
-from launch_client import schemas  # noqa: F401
+
+from launch.api_client import schemas  # noqa: F401
 
 class CloneModelBundleV1Request(schemas.DictSchema):
     """NOTE: This class is auto generated by OpenAPI Generator.

launch/api_client/model/clone_model_bundle_v2_request.pyi

@@ -19,7 +19,8 @@ from datetime import date, datetime  # noqa: F401
 
 import frozendict  # noqa: F401
 import typing_extensions  # noqa: F401
-from launch_client import schemas  # noqa: F401
+
+from launch.api_client import schemas  # noqa: F401
 
 class CloneModelBundleV2Request(schemas.DictSchema):
     """NOTE: This class is auto generated by OpenAPI Generator.

launch/api_client/model/cloudpickle_artifact_flavor.pyi

@@ -19,7 +19,8 @@ from datetime import date, datetime  # noqa: F401
 
 import frozendict  # noqa: F401
 import typing_extensions  # noqa: F401
-from launch_client import schemas  # noqa: F401
+
+from launch.api_client import schemas  # noqa: F401
 
 class CloudpickleArtifactFlavor(schemas.DictSchema):
     """NOTE: This class is auto generated by OpenAPI Generator.
@@ -310,6 +311,6 @@ class CloudpickleArtifactFlavor(schemas.DictSchema):
             **kwargs,
         )
 
-from launch_client.model.custom_framework import CustomFramework
-from launch_client.model.pytorch_framework import PytorchFramework
-from launch_client.model.tensorflow_framework import TensorflowFramework
+from launch.api_client.model.custom_framework import CustomFramework
+from launch.api_client.model.pytorch_framework import PytorchFramework
+from launch.api_client.model.tensorflow_framework import TensorflowFramework

launch/api_client/model/create_async_task_v1_response.pyi

@@ -19,7 +19,8 @@ from datetime import date, datetime  # noqa: F401
 
 import frozendict  # noqa: F401
 import typing_extensions  # noqa: F401
-from launch_client import schemas  # noqa: F401
+
+from launch.api_client import schemas  # noqa: F401
 
 class CreateAsyncTaskV1Response(schemas.DictSchema):
     """NOTE: This class is auto generated by OpenAPI Generator.