zenml-nightly 0.55.0.dev20240124__py3-none-any.whl → 0.72.0.dev20250115__py3-none-any.whl

zenml/integrations/huggingface/steps/accelerate_runner.py

@@ -0,0 +1,166 @@
+# Apache Software License 2.0
+#
+# Copyright (c) ZenML GmbH 2024. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+"""Step function to run any ZenML step using Accelerate."""
+
+import functools
+from typing import Any, Callable, Dict, Optional, TypeVar, Union, cast
+
+import cloudpickle as pickle
+from accelerate.commands.launch import (
+    launch_command,
+    launch_command_parser,
+)
+
+from zenml import get_pipeline_context
+from zenml.logger import get_logger
+from zenml.steps import BaseStep
+from zenml.utils.function_utils import _cli_arg_name, create_cli_wrapped_script
+
+logger = get_logger(__name__)
+F = TypeVar("F", bound=Callable[..., Any])
+
+
+def run_with_accelerate(
+    step_function_top_level: Optional[BaseStep] = None,
+    **accelerate_launch_kwargs: Any,
+) -> Union[Callable[[BaseStep], BaseStep], BaseStep]:
+    """Run a function with accelerate.
+
+    Accelerate package: https://huggingface.co/docs/accelerate/en/index
+    Example:
+        ```python
+        from zenml import step, pipeline
+        from zenml.integrations.hugginface.steps import run_with_accelerate
+
+        @run_with_accelerate(num_processes=4, multi_gpu=True)
+        @step
+        def training_step(some_param: int, ...):
+            # your training code is below
+            ...
+
+        @pipeline
+        def training_pipeline(some_param: int, ...):
+            training_step(some_param, ...)
+        ```
+
+    Args:
+        step_function_top_level: The step function to run with accelerate [optional].
+            Used in functional calls like `run_with_accelerate(some_func,foo=bar)()`.
+        accelerate_launch_kwargs: A dictionary of arguments to pass along to the
+            `accelerate launch` command, including hardware selection, resource
+            allocation, and training paradigm options. Visit
+            https://huggingface.co/docs/accelerate/en/package_reference/cli#accelerate-launch
+            for more details.
+
+    Returns:
+        The accelerate-enabled version of the step.
+    """
+
+    def _decorator(step_function: BaseStep) -> BaseStep:
+        def _wrapper(
+            entrypoint: F, accelerate_launch_kwargs: Dict[str, Any]
+        ) -> F:
+            @functools.wraps(entrypoint)
+            def inner(*args: Any, **kwargs: Any) -> Any:
+                if args:
+                    raise ValueError(
+                        "Accelerated steps do not support positional arguments."
+                    )
+
+                with create_cli_wrapped_script(
+                    entrypoint, flavor="accelerate"
+                ) as (
+                    script_path,
+                    output_path,
+                ):
+                    commands = [str(script_path.absolute())]
+                    for k, v in kwargs.items():
+                        k = _cli_arg_name(k)
+                        if isinstance(v, bool):
+                            if v:
+                                commands.append(f"--{k}")
+                        elif type(v) in (list, tuple, set):
+                            for each in v:
+                                commands += [f"--{k}", f"{each}"]
+                        else:
+                            commands += [f"--{k}", f"{v}"]
+                    logger.debug(commands)
+
+                    parser = launch_command_parser()
+                    args = parser.parse_args(commands)
+                    for k, v in accelerate_launch_kwargs.items():
+                        if k in args:
+                            setattr(args, k, v)
+                        else:
+                            logger.warning(
+                                f"You passed in `{k}` as an `accelerate launch` argument, but it was not accepted. "
+                                "Please check https://huggingface.co/docs/accelerate/en/package_reference/cli#accelerate-launch "
+                                "to find out more about supported arguments and retry."
+                            )
+                    try:
+                        launch_command(args)
+                    except Exception as e:
+                        logger.error(
+                            "Accelerate training job failed... See error message for details."
+                        )
+                        raise RuntimeError(
+                            "Accelerate training job failed."
+                        ) from e
+                    else:
+                        logger.info(
+                            "Accelerate training job finished successfully."
+                        )
+                        return pickle.load(open(output_path, "rb"))
+
+            return cast(F, inner)
+
+        try:
+            get_pipeline_context()
+        except RuntimeError:
+            pass
+        else:
+            raise RuntimeError(
+                f"`{run_with_accelerate.__name__}` decorator cannot be used "
+                "in a functional way with steps, please apply decoration "
+                "directly to a step instead. This behavior will be also "
+                "allowed in future, but now it faces technical limitations.\n"
+                "Example (allowed):\n"
+                f"@{run_with_accelerate.__name__}(...)\n"
+                f"def {step_function.name}(...):\n"
+                "    ...\n"
+                "Example (not allowed):\n"
+                "def my_pipeline(...):\n"
+                f"    run_with_accelerate({step_function.name},...)(...)\n"
+            )
+
+        setattr(
+            step_function, "unwrapped_entrypoint", step_function.entrypoint
+        )
+        setattr(
+            step_function,
+            "entrypoint",
+            _wrapper(
+                step_function.entrypoint,
+                accelerate_launch_kwargs=accelerate_launch_kwargs,
+            ),
+        )
+
+        return step_function
+
+    if step_function_top_level:
+        return _decorator(step_function_top_level)
+    return _decorator

zenml/integrations/huggingface/steps/huggingface_deployer.py

@@ -0,0 +1,110 @@
+#  Copyright (c) ZenML GmbH 2024. All Rights Reserved.
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at:
+#
+#       https://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express
+#  or implied. See the License for the specific language governing
+#  permissions and limitations under the License.
+"""Implementation of the Hugging Face Deployer step."""
+
+from typing import cast
+
+from zenml import get_step_context, step
+from zenml.integrations.huggingface.model_deployers.huggingface_model_deployer import (
+    HuggingFaceModelDeployer,
+)
+from zenml.integrations.huggingface.services.huggingface_deployment import (
+    HuggingFaceDeploymentService,
+    HuggingFaceServiceConfig,
+)
+from zenml.logger import get_logger
+from zenml.model_deployers.base_model_deployer import (
+    DEFAULT_DEPLOYMENT_START_STOP_TIMEOUT,
+)
+
+logger = get_logger(__name__)
+
+
+@step(enable_cache=False)
+def huggingface_model_deployer_step(
+    service_config: HuggingFaceServiceConfig,
+    deploy_decision: bool = True,
+    timeout: int = DEFAULT_DEPLOYMENT_START_STOP_TIMEOUT,
+) -> HuggingFaceDeploymentService:
+    """Hugging Face model deployer pipeline step.
+
+    This step can be used in a pipeline to implement continuous
+    deployment with Hugging Face Inference Endpoint.
+
+    Args:
+        service_config: Hugging Face deployment service configuration.
+        deploy_decision: whether to deploy the model or not
+        timeout: the timeout in seconds to wait for the deployment to start
+
+    Returns:
+        Huggingface deployment service
+    """
+    model_deployer = cast(
+        HuggingFaceModelDeployer,
+        HuggingFaceModelDeployer.get_active_model_deployer(),
+    )
+
+    # get pipeline name, step name and run id
+    context = get_step_context()
+    pipeline_name = context.pipeline.name
+    step_name = context.step_run.name
+
+    # update the step configuration with the real pipeline runtime information
+    service_config = service_config.model_copy()
+    service_config.pipeline_name = pipeline_name
+    service_config.pipeline_step_name = step_name
+
+    # fetch existing services with same pipeline name, step name and
+    # model name
+    existing_services = model_deployer.find_model_server(
+        config=service_config.model_dump()
+    )
+
+    # even when the deploy decision is negative, if an existing model server
+    # is not running for this pipeline/step, we still have to serve the
+    # current model, to ensure that a model server is available at all times
+    if not deploy_decision and existing_services:
+        logger.info(
+            f"Skipping model deployment because the model quality does not "
+            f"meet the criteria. Reusing last model server deployed by step "
+            f"'{step_name}' and pipeline '{pipeline_name}' for model "
+            f"'{service_config.model_name}'..."
+        )
+        service = cast(HuggingFaceDeploymentService, existing_services[0])
+        # even when the deploy decision is negative, we still need to start
+        # the previous model server if it is no longer running, to ensure that
+        # a model server is available at all times
+        if not service.is_running:
+            service.start(timeout=timeout)
+        return service
+
+    # invoke the Hugging Face model deployer to create a new service
+    # or update an existing one that was previously deployed for the same
+    # model
+    service = cast(
+        HuggingFaceDeploymentService,
+        model_deployer.deploy_model(
+            service_config,
+            replace=True,
+            timeout=timeout,
+            service_type=HuggingFaceDeploymentService.SERVICE_TYPE,
+        ),
+    )
+
+    logger.info(
+        f"Hugging Face deployment service started and reachable at:\n"
+        f"    {service.prediction_url}\n"
+    )
+
+    return service

zenml/integrations/hyperai/__init__.py

@@ -0,0 +1,53 @@
+#  Copyright (c) ZenML GmbH 2024. All Rights Reserved.
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at:
+#
+#       https://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express
+#  or implied. See the License for the specific language governing
+#  permissions and limitations under the License.
+"""Initialization of the HyperAI integration."""
+from typing import List, Type
+
+from zenml.integrations.constants import HYPERAI
+from zenml.integrations.integration import Integration
+from zenml.stack import Flavor
+
+# Service connector constants
+HYPERAI_CONNECTOR_TYPE = "hyperai"
+HYPERAI_RESOURCE_TYPE = "hyperai-instance"
+
+
+class HyperAIIntegration(Integration):
+    """Definition of HyperAI integration for ZenML."""
+
+    NAME = HYPERAI
+    REQUIREMENTS = [
+        "paramiko>=3.4.0",
+    ]
+
+    @classmethod
+    def activate(cls) -> None:
+        """Activates the integration."""
+        from zenml.integrations.hyperai import service_connectors  # noqa
+
+    @classmethod
+    def flavors(cls) -> List[Type[Flavor]]:
+        """Declare the stack component flavors for the HyperAI integration.
+        
+        Returns:
+            List of stack component flavors for this integration.
+        """
+        from zenml.integrations.hyperai.flavors import (
+            HyperAIOrchestratorFlavor
+        )
+
+        return [HyperAIOrchestratorFlavor]
+
+
+HyperAIIntegration.check_installation()

zenml/integrations/hyperai/flavors/__init__.py

@@ -0,0 +1,20 @@
+#  Copyright (c) ZenML GmbH 2024. All Rights Reserved.
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at:
+#
+#       https://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express
+#  or implied. See the License for the specific language governing
+#  permissions and limitations under the License.
+"""HyperAI integration flavors."""
+
+from zenml.integrations.hyperai.flavors.hyperai_orchestrator_flavor import (
+    HyperAIOrchestratorFlavor,
+)
+
+__all__ = ["HyperAIOrchestratorFlavor"]

zenml/integrations/hyperai/flavors/hyperai_orchestrator_flavor.py

@@ -0,0 +1,170 @@
+#  Copyright (c) ZenML GmbH 2024. All Rights Reserved.
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at:
+#
+#       https://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express
+#  or implied. See the License for the specific language governing
+#  permissions and limitations under the License.
+"""Implementation of the ZenML HyperAI orchestrator."""
+
+from typing import TYPE_CHECKING, Dict, Optional, Type
+
+from zenml.config.base_settings import BaseSettings
+from zenml.integrations.hyperai import HYPERAI_RESOURCE_TYPE
+from zenml.logger import get_logger
+from zenml.models import ServiceConnectorRequirements
+from zenml.orchestrators import (
+    BaseOrchestratorConfig,
+    BaseOrchestratorFlavor,
+)
+
+if TYPE_CHECKING:
+    from zenml.integrations.hyperai.orchestrators import HyperAIOrchestrator
+
+logger = get_logger(__name__)
+
+
+class HyperAIOrchestratorSettings(BaseSettings):
+    """HyperAI orchestrator settings.
+
+    Attributes:
+        mounts_from_to: A dictionary mapping from paths on the HyperAI instance
+            to paths within the Docker container. This allows users to mount
+            directories from the HyperAI instance into the Docker container that runs
+            on it.
+    """
+
+    mounts_from_to: Dict[str, str] = {}
+
+
+class HyperAIOrchestratorConfig(
+    BaseOrchestratorConfig, HyperAIOrchestratorSettings
+):
+    """Configuration for the HyperAI orchestrator.
+
+    Attributes:
+        container_registry_autologin: If True, the orchestrator will attempt to
+            automatically log in to the container registry specified in the stack
+            configuration on the HyperAI instance. This is useful if the container
+            registry requires authentication and the HyperAI instance has not been
+            manually logged in to the container registry. Defaults to `False`.
+        automatic_cleanup_pipeline_files: If True, the orchestrator will
+            automatically clean up old pipeline files that are on the HyperAI
+            instance. Pipeline files will be cleaned up if they are 7 days old or
+            older. Defaults to `True`.
+        gpu_enabled_in_container: If True, the orchestrator will enable GPU
+            support in the Docker container that runs on the HyperAI instance.
+            Defaults to `True`.
+
+    """
+
+    container_registry_autologin: bool = False
+    automatic_cleanup_pipeline_files: bool = True
+    gpu_enabled_in_container: bool = True
+
+    @property
+    def is_remote(self) -> bool:
+        """Checks if this stack component is running remotely.
+
+        This designation is used to determine if the stack component can be
+        used with a local ZenML database or if it requires a remote ZenML
+        server.
+
+        Returns:
+            True if this config is for a remote component, False otherwise.
+        """
+        return True
+
+    @property
+    def is_schedulable(self) -> bool:
+        """Whether the orchestrator is schedulable or not.
+
+        Returns:
+            Whether the orchestrator is schedulable or not.
+        """
+        return True
+
+
+class HyperAIOrchestratorFlavor(BaseOrchestratorFlavor):
+    """Flavor for the HyperAI orchestrator."""
+
+    @property
+    def name(self) -> str:
+        """Name of the orchestrator flavor.
+
+        Returns:
+            Name of the orchestrator flavor.
+        """
+        return "hyperai"
+
+    @property
+    def service_connector_requirements(
+        self,
+    ) -> Optional[ServiceConnectorRequirements]:
+        """Service connector resource requirements for service connectors.
+
+        Specifies resource requirements that are used to filter the available
+        service connector types that are compatible with this flavor.
+
+        Returns:
+            Requirements for compatible service connectors, if a service
+            connector is required for this flavor.
+        """
+        return ServiceConnectorRequirements(
+            resource_type=HYPERAI_RESOURCE_TYPE
+        )
+
+    @property
+    def docs_url(self) -> Optional[str]:
+        """A url to point at docs explaining this flavor.
+
+        Returns:
+            A flavor docs url.
+        """
+        return self.generate_default_docs_url()
+
+    @property
+    def sdk_docs_url(self) -> Optional[str]:
+        """A url to point at SDK docs explaining this flavor.
+
+        Returns:
+            A flavor SDK docs url.
+        """
+        return self.generate_default_sdk_docs_url()
+
+    @property
+    def logo_url(self) -> str:
+        """A url to represent the flavor in the dashboard.
+
+        Returns:
+            The flavor logo.
+        """
+        return "https://public-flavor-logos.s3.eu-central-1.amazonaws.com/connectors/hyperai/hyperai.png"
+
+    @property
+    def config_class(self) -> Type[BaseOrchestratorConfig]:
+        """Config class for the base orchestrator flavor.
+
+        Returns:
+            The config class.
+        """
+        return HyperAIOrchestratorConfig
+
+    @property
+    def implementation_class(self) -> Type["HyperAIOrchestrator"]:
+        """Implementation class for this flavor.
+
+        Returns:
+            Implementation class for this flavor.
+        """
+        from zenml.integrations.hyperai.orchestrators import (
+            HyperAIOrchestrator,
+        )
+
+        return HyperAIOrchestrator

zenml/integrations/hyperai/orchestrators/__init__.py

@@ -0,0 +1,21 @@
+#  Copyright (c) ZenML GmbH 2024. All Rights Reserved.
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at:
+#
+#       https://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express
+#  or implied. See the License for the specific language governing
+#  permissions and limitations under the License.
+"""HyperAI orchestrator."""
+from zenml.integrations.hyperai.orchestrators.hyperai_orchestrator import (
+    HyperAIOrchestrator
+)
+
+__all__ = [
+    "HyperAIOrchestrator"
+]