[Extended Resources] GPU Accelerators

flytekit/core/base_task.py

@@ -24,6 +24,8 @@
 from dataclasses import dataclass
 from typing import Any, Coroutine, Dict, Generic, List, Optional, OrderedDict, Tuple, Type, TypeVar, Union, cast
 
+from flyteidl.core import tasks_pb2
+
 from flytekit.configuration import SerializationSettings
 from flytekit.core.context_manager import (
     ExecutionParameters,
@@ -344,6 +346,12 @@
         """
         return None
 
+    def get_extended_resources(self, settings: SerializationSettings) -> Optional[tasks_pb2.ExtendedResources]:
+        """
+        Returns the extended resources to allocate to the task on hosted Flyte.
+        """
+        return None
+
     def local_execution_mode(self) -> ExecutionState.Mode:
         """ """
         return ExecutionState.Mode.LOCAL_TASK_EXECUTION

flytekit/core/node.py

@@ -4,6 +4,8 @@
 import typing
 from typing import Any, List
 
+from flyteidl.core import tasks_pb2
+
 from flytekit.core.resources import Resources, convert_resources_to_resource_model
 from flytekit.core.utils import _dnsify
 from flytekit.loggers import logger
@@ -62,6 +64,7 @@
         self._aliases: _workflow_model.Alias = None
         self._outputs = None
         self._resources: typing.Optional[_resources_model] = None
+        self._extended_resources: typing.Optional[tasks_pb2.ExtendedResources] = None
 
     def runs_before(self, other: Node):
         """
@@ -172,6 +175,11 @@
             assert_not_promise(v, "container_image")
             self.flyte_entity._container_image = v
 
+        if "accelerator" in kwargs:
+            v = kwargs["accelerator"]
+            assert_not_promise(v, "accelerator")
+            self._extended_resources = tasks_pb2.ExtendedResources(gpu_accelerator=v.to_flyte_idl())
+
         return self
 
 

flytekit/core/python_auto_container.py

@@ -5,6 +5,8 @@
 from abc import ABC
 from typing import Callable, Dict, List, Optional, TypeVar, Union
 
+from flyteidl.core import tasks_pb2
+
 from flytekit.configuration import ImageConfig, SerializationSettings
 from flytekit.core.base_task import PythonTask, TaskMetadata, TaskResolverMixin
 from flytekit.core.context_manager import FlyteContextManager
@@ -13,6 +15,7 @@
 from flytekit.core.tracked_abc import FlyteTrackedABC
 from flytekit.core.tracker import TrackedInstance, extract_task_module
 from flytekit.core.utils import _get_container_definition, _serialize_pod_spec, timeit
+from flytekit.extras.accelerators import BaseAccelerator
 from flytekit.image_spec.image_spec import ImageBuildEngine, ImageSpec
 from flytekit.loggers import logger
 from flytekit.models import task as _task_model
@@ -44,6 +47,7 @@
         secret_requests: Optional[List[Secret]] = None,
         pod_template: Optional[PodTemplate] = None,
         pod_template_name: Optional[str] = None,
+        accelerator: Optional[BaseAccelerator] = None,
         **kwargs,
     ):
         """
@@ -70,6 +74,7 @@
            - `AWS Parameter store <https://docs.aws.amazon.com/systems-manager/latest/userguide/systems-manager-parameter-store.html>`__
         :param pod_template: Custom PodTemplate for this task.
         :param pod_template_name: The name of the existing PodTemplate resource which will be used in this task.
+        :param accelerator: The accelerator to use for this task.
         """
         sec_ctx = None
         if secret_requests:
@@ -110,6 +115,7 @@
         self._get_command_fn = self.get_default_command
 
         self.pod_template = pod_template
+        self.accelerator = accelerator
 
     @property
     def task_resolver(self) -> TaskResolverMixin:
@@ -219,6 +225,15 @@
             return {}
         return {_PRIMARY_CONTAINER_NAME_FIELD: self.pod_template.primary_container_name}
 
+    def get_extended_resources(self, settings: SerializationSettings) -> Optional[tasks_pb2.ExtendedResources]:
+        """
+        Returns the extended resources to allocate to the task on hosted Flyte.
+        """
+        if self.accelerator is None:
+            return None
+
+        return tasks_pb2.ExtendedResources(gpu_accelerator=self.accelerator.to_flyte_idl())
+
 
 class DefaultTaskResolver(TrackedInstance, TaskResolverMixin):
     """

flytekit/core/task.py

@@ -8,6 +8,7 @@
 from flytekit.core.python_function_task import PythonFunctionTask
 from flytekit.core.reference_entity import ReferenceEntity, TaskReference
 from flytekit.core.resources import Resources
+from flytekit.extras.accelerators import BaseAccelerator
 from flytekit.image_spec.image_spec import ImageSpec
 from flytekit.models.documentation import Documentation
 from flytekit.models.security import Secret
@@ -102,6 +103,7 @@ def task(
     enable_deck: Optional[bool] = ...,
     pod_template: Optional["PodTemplate"] = ...,
     pod_template_name: Optional[str] = ...,
+    accelerator: Optional[BaseAccelerator] = ...,
 ) -> Callable[[Callable[..., FuncOut]], PythonFunctionTask[T]]:
     ...
 
@@ -129,6 +131,7 @@ def task(
     enable_deck: Optional[bool] = ...,
     pod_template: Optional["PodTemplate"] = ...,
     pod_template_name: Optional[str] = ...,
+    accelerator: Optional[BaseAccelerator] = ...,
 ) -> Union[PythonFunctionTask[T], Callable[..., FuncOut]]:
     ...
 
@@ -155,6 +158,7 @@ def task(
     enable_deck: Optional[bool] = None,
     pod_template: Optional["PodTemplate"] = None,
     pod_template_name: Optional[str] = None,
+    accelerator: Optional[BaseAccelerator] = None,
 ) -> Union[Callable[[Callable[..., FuncOut]], PythonFunctionTask[T]], PythonFunctionTask[T], Callable[..., FuncOut]]:
     """
     This is the core decorator to use for any task type in flytekit.
@@ -248,6 +252,7 @@ def foo2():
     :param docs: Documentation about this task
     :param pod_template: Custom PodTemplate for this task.
     :param pod_template_name: The name of the existing PodTemplate resource which will be used in this task.
+    :param accelerator: The accelerator to use for this task.
     """
 
     def wrapper(fn: Callable[..., Any]) -> PythonFunctionTask[T]:
@@ -277,6 +282,7 @@ def wrapper(fn: Callable[..., Any]) -> PythonFunctionTask[T]:
             docs=docs,
             pod_template=pod_template,
             pod_template_name=pod_template_name,
+            accelerator=accelerator,
         )
         update_wrapper(task_instance, fn)
         return task_instance

flytekit/extras/accelerators.py

@@ -0,0 +1,90 @@
+import abc
+import copy
+from typing import ClassVar, Generic, Optional, Type, TypeVar
+
+from flyteidl.core import tasks_pb2
+
+T = TypeVar("T")
+MIG = TypeVar("MIG", bound="MultiInstanceGPUAccelerator")
+
+
+class BaseAccelerator(abc.ABC, Generic[T]):
+    @abc.abstractmethod
+    def to_flyte_idl(self) -> T:
+        ...
+
+
+class GPUAccelerator(BaseAccelerator):
+    def __init__(self, device: str) -> None:
+        self._device = device
+
+    def to_flyte_idl(self) -> tasks_pb2.GPUAccelerator:
+        return tasks_pb2.GPUAccelerator(device=self._device)
+
+
+A10G = GPUAccelerator("nvidia-a10g")
+L4 = GPUAccelerator("nvidia-l4-vws")
+K80 = GPUAccelerator("nvidia-tesla-k80")
+M60 = GPUAccelerator("nvidia-tesla-m60")
+P4 = GPUAccelerator("nvidia-tesla-p4")
+P100 = GPUAccelerator("nvidia-tesla-p100")
+T4 = GPUAccelerator("nvidia-tesla-t4")
+V100 = GPUAccelerator("nvidia-tesla-v100")
+
+
+class MultiInstanceGPUAccelerator(BaseAccelerator):
+    device: ClassVar[str]
+    _partition_size: Optional[str]
+
+    @property
+    def unpartitioned(self: MIG) -> MIG:
+        instance = copy.deepcopy(self)
+        instance._partition_size = None
+        return instance
+
+    @classmethod
+    def partitioned(cls: Type[MIG], partition_size: str) -> MIG:
+        instance = cls()
+        instance._partition_size = partition_size
+        return instance
+
+    def to_flyte_idl(self) -> tasks_pb2.GPUAccelerator:
+        msg = tasks_pb2.GPUAccelerator(device=self.device)
+        if not hasattr(self, "_partition_size"):
+            return msg
+
+        if self._partition_size is None:
+            msg.unpartitioned = True
+        else:
+            msg.partition_size = self._partition_size
+        return msg
+
+
+class _A100_Base(MultiInstanceGPUAccelerator):
+    device = "nvidia-tesla-a100"
+
+
+class _A100(_A100_Base):
+    partition_1g_5gb = _A100_Base.partitioned("1g.5gb")
+    partition_2g_10gb = _A100_Base.partitioned("2g.10gb")
+    partition_3g_20gb = _A100_Base.partitioned("3g.20gb")
+    partition_4g_20gb = _A100_Base.partitioned("4g.20gb")
+    partition_7g_40gb = _A100_Base.partitioned("7g.40gb")
+
+
+A100 = _A100()
+
+
+class _A100_80GB_Base(MultiInstanceGPUAccelerator):
+    device = "nvidia-a100-80gb"
+
+
+class _A100_80GB(_A100_80GB_Base):
+    partition_1g_10gb = _A100_80GB_Base.partitioned("1g.10gb")
+    partition_2g_20gb = _A100_80GB_Base.partitioned("2g.20gb")
+    partition_3g_40gb = _A100_80GB_Base.partitioned("3g.40gb")
+    partition_4g_40gb = _A100_80GB_Base.partitioned("4g.40gb")
+    partition_7g_80gb = _A100_80GB_Base.partitioned("7g.80gb")
+
+
+A100_80GB = _A100_80GB()

flytekit/models/core/workflow.py

@@ -1,6 +1,7 @@
 import datetime
 import typing
 
+from flyteidl.core import tasks_pb2
 from flyteidl.core import workflow_pb2 as _core_workflow
 
 from flytekit.models import common as _common
@@ -562,24 +563,33 @@
 
 
 class TaskNodeOverrides(_common.FlyteIdlEntity):
-    def __init__(self, resources: typing.Optional[Resources] = None):
+    def __init__(
+        self, resources: typing.Optional[Resources], extended_resources: typing.Optional[tasks_pb2.ExtendedResources]
+    ):
         self._resources = resources
+        self._extended_resources = extended_resources
 
     @property
     def resources(self) -> Resources:
         return self._resources
 
+    @property
+    def extended_resources(self) -> tasks_pb2.ExtendedResources:
+        return self._extended_resources
+
     def to_flyte_idl(self):
         return _core_workflow.TaskNodeOverrides(
             resources=self.resources.to_flyte_idl() if self.resources is not None else None,
+            extended_resources=self.extended_resources,
         )
 
     @classmethod
     def from_flyte_idl(cls, pb2_object):
         resources = Resources.from_flyte_idl(pb2_object.resources)
+        extended_resources = pb2_object.extended_resources if pb2_object.HasField("extended_resources") else None
         if bool(resources.requests) or bool(resources.limits):
-            return cls(resources=resources)
-        return cls(resources=None)
+            return cls(resources=resources, extended_resources=extended_resources)
+        return cls(resources=None, extended_resources=extended_resources)
 
 
 class TaskNode(_common.FlyteIdlEntity):

flytekit/models/task.py

@@ -336,6 +336,7 @@
         config=None,
         k8s_pod=None,
         sql=None,
+        extended_resources=None,
     ):
         """
         A task template represents the full set of information necessary to perform a unit of work in the Flyte system.
@@ -359,6 +360,7 @@
             in tandem with the custom.
         :param K8sPod k8s_pod: Alternative to the container used to execute this task.
         :param Sql sql: This is used to execute query in FlytePropeller instead of running container or k8s_pod.
+        :param flyteidl.core.tasks_pb2.ExtendedResources extended_resources: The extended resources to allocate to the task.
         """
         if (
             (container is not None and k8s_pod is not None)
@@ -377,6 +379,7 @@
         self._security_context = security_context
         self._k8s_pod = k8s_pod
         self._sql = sql
+        self._extended_resources = extended_resources
 
     @property
     def id(self):
@@ -451,6 +454,14 @@
     def sql(self):
         return self._sql
 
+    @property
+    def extended_resources(self):
+        """
+        If not None, the extended resources to allocate to the task.
+        :rtype: flyteidl.core.tasks_pb2.ExtendedResources
+        """
+        return self._extended_resources
+
     def to_flyte_idl(self):
         """
         :rtype: flyteidl.core.tasks_pb2.TaskTemplate
@@ -464,6 +475,7 @@
             container=self.container.to_flyte_idl() if self.container else None,
             task_type_version=self.task_type_version,
             security_context=self.security_context.to_flyte_idl() if self.security_context else None,
+            extended_resources=self.extended_resources,
             config={k: v for k, v in self.config.items()} if self.config is not None else None,
             k8s_pod=self.k8s_pod.to_flyte_idl() if self.k8s_pod else None,
             sql=self.sql.to_flyte_idl() if self.sql else None,
@@ -487,6 +499,7 @@
             security_context=_sec.SecurityContext.from_flyte_idl(pb2_object.security_context)
             if pb2_object.security_context and pb2_object.security_context.ByteSize() > 0
             else None,
+            extended_resources=pb2_object.extended_resources if pb2_object.HasField("extended_resources") else None,
             config={k: v for k, v in pb2_object.config.items()} if pb2_object.config is not None else None,
             k8s_pod=K8sPod.from_flyte_idl(pb2_object.k8s_pod) if pb2_object.HasField("k8s_pod") else None,
             sql=Sql.from_flyte_idl(pb2_object.sql) if pb2_object.HasField("sql") else None,

flytekit/tools/translator.py

@@ -214,6 +214,7 @@ def get_serializable_task(
         config=entity.get_config(settings),
         k8s_pod=pod,
         sql=entity.get_sql(settings),
+        extended_resources=entity.get_extended_resources(settings),
     )
     if settings.should_fast_serialize() and isinstance(entity, PythonAutoContainerTask):
         entity.reset_command_fn()
@@ -440,7 +441,8 @@ def get_serializable_node(
             upstream_node_ids=[n.id for n in upstream_nodes],
             output_aliases=[],
             task_node=workflow_model.TaskNode(
-                reference_id=task_spec.template.id, overrides=TaskNodeOverrides(resources=entity._resources)
+                reference_id=task_spec.template.id,
+                overrides=TaskNodeOverrides(resources=entity._resources, extended_resources=entity._extended_resources),
             ),
         )
         if entity._aliases:
@@ -516,7 +518,8 @@ def get_serializable_node(
             upstream_node_ids=[n.id for n in upstream_nodes],
             output_aliases=[],
             task_node=workflow_model.TaskNode(
-                reference_id=entity.flyte_entity.id, overrides=TaskNodeOverrides(resources=entity._resources)
+                reference_id=entity.flyte_entity.id,
+                overrides=TaskNodeOverrides(resources=entity._resources, extended_resources=entity._extended_resources),
             ),
         )
     elif isinstance(entity.flyte_entity, FlyteWorkflow):
@@ -565,7 +568,7 @@ def get_serializable_array_node(
     task_spec = get_serializable(entity_mapping, settings, entity, options)
     task_node = workflow_model.TaskNode(
         reference_id=task_spec.template.id,
-        overrides=TaskNodeOverrides(resources=node._resources),
+        overrides=TaskNodeOverrides(resources=node._resources, extended_resources=node._extended_resources),
     )
     node = workflow_model.Node(
         id=entity.name,