nvidia-nat 1.2.0__py3-none-any.whl

nat/front_ends/fastapi/fastapi_front_end_config.py

@@ -0,0 +1,241 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# 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.
+
+import logging
+import typing
+from datetime import datetime
+from pathlib import Path
+
+from pydantic import BaseModel
+from pydantic import Field
+from pydantic import field_validator
+
+from nat.data_models.component_ref import ObjectStoreRef
+from nat.data_models.front_end import FrontEndBaseConfig
+from nat.data_models.step_adaptor import StepAdaptorConfig
+
+logger = logging.getLogger(__name__)
+
+YAML_EXTENSIONS = (".yaml", ".yml")
+
+
+class EvaluateRequest(BaseModel):
+    """Request model for the evaluate endpoint."""
+    config_file: str = Field(description="Path to the configuration file for evaluation")
+    job_id: str | None = Field(default=None, description="Unique identifier for the evaluation job")
+    reps: int = Field(default=1, gt=0, description="Number of repetitions for the evaluation, defaults to 1")
+    expiry_seconds: int = Field(
+        default=3600,
+        gt=0,
+        description="Optional time (in seconds) before the job expires. Clamped between 600 (10 min) and 86400 (24h).")
+
+    @field_validator('job_id', mode='after')
+    @classmethod
+    def validate_job_id(cls, job_id: str):
+        job_id = job_id.strip()
+        job_id_path = Path(job_id)
+        if len(job_id_path.parts) > 1 or job_id_path.resolve().name != job_id:
+            raise ValueError(
+                f"Job ID '{job_id}' contains invalid characters. Only alphanumeric characters and underscores are"
+                " allowed.")
+
+        if job_id_path.is_reserved():
+            # reserved names is Windows specific
+            raise ValueError(f"Job ID '{job_id}' is a reserved name. Please choose a different name.")
+
+        return job_id
+
+    @field_validator('config_file', mode='after')
+    @classmethod
+    def validate_config_file(cls, config_file: str):
+        config_file = config_file.strip()
+        config_file_path = Path(config_file).resolve()
+
+        # Ensure the config file is a YAML file
+        if config_file_path.suffix.lower() not in YAML_EXTENSIONS:
+            raise ValueError(f"Config file '{config_file}' must be a YAML file with one of the following extensions: "
+                             f"{', '.join(YAML_EXTENSIONS)}")
+
+        if config_file_path.is_reserved():
+            # reserved names is Windows specific
+            raise ValueError(f"Config file '{config_file}' is a reserved name. Please choose a different name.")
+
+        if not config_file_path.exists():
+            raise ValueError(f"Config file '{config_file}' does not exist. Please provide a valid path.")
+
+        return config_file
+
+
+class BaseAsyncResponse(BaseModel):
+    """Base model for async responses."""
+    job_id: str = Field(description="Unique identifier for the job")
+    status: str = Field(description="Current status of the job")
+
+
+class EvaluateResponse(BaseAsyncResponse):
+    """Response model for the evaluate endpoint."""
+    pass
+
+
+class AsyncGenerateResponse(BaseAsyncResponse):
+    """Response model for the async generation endpoint."""
+    pass
+
+
+class BaseAsyncStatusResponse(BaseModel):
+    """Base model for async status responses."""
+    job_id: str = Field(description="Unique identifier for the evaluation job")
+    status: str = Field(description="Current status of the evaluation job")
+    error: str | None = Field(default=None, description="Error message if the job failed")
+    created_at: datetime = Field(description="Timestamp when the job was created")
+    updated_at: datetime = Field(description="Timestamp when the job was last updated")
+    expires_at: datetime | None = Field(default=None, description="Timestamp when the job will expire")
+
+
+class EvaluateStatusResponse(BaseAsyncStatusResponse):
+    """Response model for the evaluate status endpoint."""
+    config_file: str = Field(description="Path to the configuration file used for evaluation")
+    output_path: str | None = Field(default=None,
+                                    description="Path to the output file if the job completed successfully")
+
+
+class AsyncGenerationStatusResponse(BaseAsyncStatusResponse):
+    output: dict | None = Field(
+        default=None,
+        description="Output of the generate request, this is only available if the job completed successfully.")
+
+
+class FastApiFrontEndConfig(FrontEndBaseConfig, name="fastapi"):
+    """
+    A FastAPI based front end that allows a NAT workflow to be served as a microservice.
+    """
+
+    class EndpointBase(BaseModel):
+
+        method: typing.Literal["GET", "POST", "PUT", "DELETE"]
+        description: str
+        path: str | None = Field(
+            default=None,
+            description=("Path for the default workflow. If None, no workflow endpoint is created."),
+        )
+        websocket_path: str | None = Field(
+            default=None,
+            description=("Path for the websocket. If None, no websocket is created."),
+        )
+        openai_api_path: str | None = Field(
+            default=None,
+            description=("Path for the default workflow using the OpenAI API Specification. "
+                         "If None, no workflow endpoint with the OpenAI API Specification is created."),
+        )
+        openai_api_v1_path: str | None = Field(
+            default=None,
+            description=("Path for the OpenAI v1 Chat Completions API compatible endpoint. "
+                         "If provided, creates a single endpoint that handles both streaming and "
+                         "non-streaming requests based on the 'stream' parameter, following the "
+                         "OpenAI Chat Completions API specification exactly."),
+        )
+
+    class Endpoint(EndpointBase):
+        function_name: str = Field(description="The name of the function to call for this endpoint")
+
+    class CrossOriginResourceSharing(BaseModel):
+        allow_origins: list[str] | None = Field(
+            default=None, description=" A list of origins that should be permitted to make cross-origin requests.")
+        allow_origin_regex: str | None = Field(
+            default=None,
+            description="A permitted regex string to match against origins to make cross-origin requests",
+        )
+        allow_methods: list[str] | None = Field(
+            default_factory=lambda: ['GET'],
+            description="A list of HTTP methods that should be allowed for cross-origin requests.")
+        allow_headers: list[str] | None = Field(
+            default_factory=list,
+            description="A list of HTTP request headers that should be supported for cross-origin requests.")
+        allow_credentials: bool | None = Field(
+            default=False,
+            description="Indicate that cookies should be supported for cross-origin requests.",
+        )
+        expose_headers: list[str] | None = Field(
+            default_factory=list,
+            description="Indicate any response headers that should be made accessible to the browser.",
+        )
+        max_age: int | None = Field(
+            default=600,
+            description="Sets a maximum time in seconds for browsers to cache CORS responses.",
+        )
+
+    root_path: str = Field(default="", description="The root path for the API")
+    host: str = Field(default="localhost", description="Host to bind the server to")
+    port: int = Field(default=8000, description="Port to bind the server to", ge=0, le=65535)
+    reload: bool = Field(default=False, description="Enable auto-reload for development")
+    workers: int = Field(default=1, description="Number of workers to run", ge=1)
+    max_running_async_jobs: int = Field(default=10,
+                                        description="Maximum number of async jobs to run concurrently",
+                                        ge=1)
+    step_adaptor: StepAdaptorConfig = StepAdaptorConfig()
+
+    workflow: typing.Annotated[EndpointBase, Field(description="Endpoint for the default workflow.")] = EndpointBase(
+        method="POST",
+        path="/generate",
+        websocket_path="/websocket",
+        openai_api_path="/chat",
+        openai_api_v1_path="/v1/chat/completions",
+        description="Executes the default NAT workflow from the loaded configuration ",
+    )
+
+    evaluate: typing.Annotated[EndpointBase, Field(description="Endpoint for evaluating workflows.")] = EndpointBase(
+        method="POST",
+        path="/evaluate",
+        description="Evaluates the performance and accuracy of the workflow on a dataset",
+    )
+
+    oauth2_callback_path: str | None = Field(
+        default="/auth/redirect",
+        description="OAuth2.0 authentication callback endpoint. If None, no OAuth2 callback endpoint is created.")
+
+    endpoints: list[Endpoint] = Field(
+        default_factory=list,
+        description=("Additional endpoints to add to the FastAPI app which run functions within the NAT configuration. "
+                     "Each endpoint must have a unique path."))
+
+    cors: CrossOriginResourceSharing = Field(
+        default_factory=CrossOriginResourceSharing,
+        description="Cross origin resource sharing configuration for the FastAPI app")
+
+    use_gunicorn: bool = Field(
+        default=False,
+        description="Use Gunicorn to run the FastAPI app",
+    )
+    runner_class: str | None = Field(
+        default=None,
+        description=("The NAT runner class to use when launching the FastAPI app from multiple processes. "
+                     "Each runner is responsible for loading and running the NAT workflow. "
+                     "Note: This is different from the worker class used by Gunicorn."),
+    )
+
+    object_store: ObjectStoreRef | None = Field(
+        default=None,
+        description=(
+            "Object store reference for the FastAPI app. If present, static files can be uploaded via a POST "
+            "request to '/static' and files will be served from the object store. The files will be served from the "
+            "object store at '/static/{file_name}'."))
+
+
+# Compatibility aliases with previous releases
+AIQEvaluateRequest = EvaluateRequest
+AIQEvaluateResponse = EvaluateResponse
+AIQAsyncGenerateResponse = AsyncGenerateResponse
+AIQEvaluateStatusResponse = EvaluateStatusResponse
+AIQAsyncGenerationStatusResponse = AsyncGenerationStatusResponse

nat/front_ends/fastapi/fastapi_front_end_controller.py

@@ -0,0 +1,68 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# 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.
+
+import asyncio
+import logging
+
+from fastapi import FastAPI
+from uvicorn import Config
+from uvicorn import Server
+
+logger = logging.getLogger(__name__)
+
+
+class _FastApiFrontEndController:
+    """
+    _FastApiFrontEndController class controls the spawing and tear down of the API server in environments where
+    the server is needed and not already running.
+    """
+
+    def __init__(self, app: FastAPI):
+        self._app: FastAPI = app
+        self._server: Server | None = None
+        self._server_background_task: asyncio.Task | None = None
+
+    async def start_server(self, host: str, port: int) -> None:
+        """Starts the API server."""
+
+        server_host = host
+        server_port = port
+
+        config = Config(app=self._app, host=server_host, port=server_port, log_level="warning")
+        self._server = Server(config=config)
+
+        try:
+            self._server_background_task = asyncio.create_task(self._server.serve())
+        except asyncio.CancelledError as e:
+            error_message = f"Task error occurred while starting API server: {str(e)}"
+            logger.error(error_message, exc_info=True)
+            raise RuntimeError(error_message) from e
+        except Exception as e:
+            error_message = f"Unexpected error occurred while starting API server: {str(e)}"
+            logger.error(error_message, exc_info=True)
+            raise RuntimeError(error_message) from e
+
+    async def stop_server(self) -> None:
+        """Stops the API server."""
+        if not self._server or not self._server_background_task:
+            return
+
+        try:
+            self._server.should_exit = True
+            await self._server_background_task
+        except asyncio.CancelledError as e:
+            logger.error("Server shutdown failed: %s", str(e), exc_info=True)
+        except Exception as e:
+            logger.error("Unexpected error occurred: %s", str(e), exc_info=True)

nat/front_ends/fastapi/fastapi_front_end_plugin.py

@@ -0,0 +1,116 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# 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.
+
+import logging
+import os
+import tempfile
+import typing
+
+from nat.builder.front_end import FrontEndBase
+from nat.front_ends.fastapi.fastapi_front_end_config import FastApiFrontEndConfig
+from nat.front_ends.fastapi.fastapi_front_end_plugin_worker import FastApiFrontEndPluginWorkerBase
+from nat.front_ends.fastapi.main import get_app
+from nat.utils.io.yaml_tools import yaml_dump
+
+logger = logging.getLogger(__name__)
+
+
+class FastApiFrontEndPlugin(FrontEndBase[FastApiFrontEndConfig]):
+
+    def get_worker_class(self) -> type[FastApiFrontEndPluginWorkerBase]:
+        from nat.front_ends.fastapi.fastapi_front_end_plugin_worker import FastApiFrontEndPluginWorker
+
+        return FastApiFrontEndPluginWorker
+
+    @typing.final
+    def get_worker_class_name(self) -> str:
+
+        if (self.front_end_config.runner_class):
+            return self.front_end_config.runner_class
+
+        worker_class = self.get_worker_class()
+
+        return f"{worker_class.__module__}.{worker_class.__qualname__}"
+
+    async def run(self):
+
+        # Write the entire config to a temporary file
+        with tempfile.NamedTemporaryFile(mode="w", prefix="nat_config", suffix=".yml", delete=False) as config_file:
+
+            # Get as dict
+            config_dict = self.full_config.model_dump(mode="json", by_alias=True, round_trip=True)
+
+            # Write to YAML file
+            yaml_dump(config_dict, config_file)
+
+            # Save the config file path for cleanup (required on Windows due to delete=False workaround)
+            config_file_name = config_file.name
+
+            # Set the config file in the environment
+            os.environ["NAT_CONFIG_FILE"] = str(config_file.name)
+
+            # Set the worker class in the environment
+            os.environ["NAT_FRONT_END_WORKER"] = self.get_worker_class_name()
+
+        try:
+            if not self.front_end_config.use_gunicorn:
+                import uvicorn
+
+                reload_excludes = ["./.*"]
+
+                uvicorn.run("nat.front_ends.fastapi.main:get_app",
+                            host=self.front_end_config.host,
+                            port=self.front_end_config.port,
+                            workers=self.front_end_config.workers,
+                            reload=self.front_end_config.reload,
+                            factory=True,
+                            reload_excludes=reload_excludes)
+
+            else:
+                app = get_app()
+
+                from gunicorn.app.wsgiapp import WSGIApplication
+
+                class StandaloneApplication(WSGIApplication):
+
+                    def __init__(self, app, options=None):
+                        self.options = options or {}
+                        self.app = app
+                        super().__init__()
+
+                    def load_config(self):
+                        config = {
+                            key: value
+                            for key, value in self.options.items() if key in self.cfg.settings and value is not None
+                        }
+                        for key, value in config.items():
+                            self.cfg.set(key.lower(), value)
+
+                    def load(self):
+                        return self.app
+
+                options = {
+                    "bind": f"{self.front_end_config.host}:{self.front_end_config.port}",
+                    "workers": self.front_end_config.workers,
+                    "worker_class": "uvicorn.workers.UvicornWorker",
+                }
+
+                StandaloneApplication(app, options=options).run()
+
+        finally:
+            try:
+                os.remove(config_file_name)
+            except OSError as e:
+                logger.error(f"Warning: Failed to delete temp file {config_file_name}: {e}")