agentscope-runtime 0.2.0b2__py3-none-any.whl → 1.0.0b1__py3-none-any.whl

agentscope_runtime/tools/base.py

@@ -0,0 +1,264 @@
+# -*- coding: utf-8 -*-
+# pylint:disable=typevar-name-incorrect-variance, unused-argument
+
+import json
+from typing import (
+    Any,
+    Dict,
+    Generic,
+    List,
+    Optional,
+    Type,
+    TypeVar,
+    Union,
+    cast,
+    get_args,
+)
+
+import jsonref
+from asgiref.sync import async_to_sync
+from pydantic import BaseModel, ValidationError
+
+from ..engine.schemas.agent_schemas import (
+    FunctionParameters,
+    FunctionTool,
+)
+
+
+# A type variable bounded by BaseModel, meaning it can represent BaseModel or
+# any subclass of it.
+ToolArgsT = TypeVar("ToolArgsT", bound=BaseModel, contravariant=True)
+ToolReturnT = TypeVar("ToolReturnT", bound=BaseModel, covariant=True)
+
+
+class Tool(Generic[ToolArgsT, ToolReturnT]):
+    """Base class for all zh, supporting both async and streaming
+    capabilities.
+    """
+
+    name: str
+    description: str
+
+    def __init__(
+        self,
+        name: Optional[str] = None,
+        description: Optional[str] = None,
+        **kwargs: Any,
+    ):
+        """Initialize the base component.
+
+        Args:
+            name: The name of the component.
+            description: The description of the component.
+            **kwargs: Other arguments if needed.
+
+        Raises:
+            ValueError: If component name and description are not provided.
+        """
+        self.name = name if name else self.__class__.name
+        self.description = (
+            description if description else self.__class__.description
+        )
+        if not self.name or not self.description:
+            raise ValueError(
+                "Tool name and description must be provided.",
+            )
+        self.input_type = self._input_type()
+        self.return_type = self._return_type()
+        self.parameters = self._parameters_parser()
+        self.function_schema = FunctionTool(
+            name=self.name,
+            description=self.description,
+            parameters=self.parameters,
+        )
+
+    async def _arun(
+        self,
+        args: ToolArgsT,
+        **kwargs: Any,
+    ) -> ToolReturnT:
+        """Actual component execution method
+
+        Args:
+            args: Input parameters adhering to the input schema.
+            **kwargs: Other arguments if needed.
+
+        Returns:
+            ToolReturnT: Output parameters adhering to the output schema.
+
+        Raises:
+            NotImplementedError: This method must be implemented by subclasses.
+        """
+        raise NotImplementedError
+
+    async def arun(
+        self,
+        args: ToolArgsT,
+        **kwargs: Any,
+    ) -> ToolReturnT:
+        """Run the component with the given arguments asynchronously.
+
+        Args:
+            args: Input parameters adhering to the input schema.
+            **kwargs: Other arguments if needed.
+
+        Returns:
+            ToolReturnT: Output parameters adhering to the output schema.
+
+        Raises:
+            TypeError: If input or return types don't match expected schemas.
+        """
+        if not isinstance(args, self.input_type):
+            raise TypeError(
+                f"The input must in the format of {self.input_type.__name__} "
+                f"or its subclass",
+            )
+
+        # make sure kwargs works
+        if not kwargs:
+            kwargs = {}
+
+        result = await self._arun(args, **kwargs)
+        if not isinstance(result, self.return_type):
+            raise TypeError(
+                f"The return must in the format of "
+                f"{self.return_type.__name__} or its subclass",
+            )
+        return result
+
+    def run(self, args: Any, **kwargs: Any) -> Any:
+        """Run the component synchronously.
+
+        Makes sure the async method could be called from sync context with or
+        without asyncio loop running.
+
+        Args:
+            args: Input arguments.
+            **kwargs: Additional keyword arguments.
+
+        Returns:
+            Any: Result of the component execution.
+        """
+        return async_to_sync(self.arun)(args, **kwargs)
+
+    def _input_type(self) -> Type[ToolArgsT]:
+        """Extract the generic input types.
+
+        Returns:
+            Type[ToolArgsT]: The input schema type,
+            used for validating input arguments.
+        """
+        return get_args(self.__orig_bases__[0])[0]
+
+    def _return_type(self) -> Type[ToolReturnT]:
+        """Extract the generic return types.
+
+        Returns:
+            Type[ToolReturnT]: The return schema type, used for
+            validating return values.
+        """
+        return get_args(self.__orig_bases__[0])[1]
+
+    def _parameters_parser(self) -> FunctionParameters:
+        """Parse the input type to generate the parameter schema.
+
+        Returns:
+            FunctionParameters: Schema representation of the input parameters.
+        """
+        try:
+            model_schema: Dict[str, Any] = self.input_type.model_json_schema()
+        except AttributeError:
+            # make sure user can  use the component without valid input type
+            return FunctionParameters(
+                type="object",
+                properties={},
+                required=[],
+            )
+
+        if "$defs" in model_schema:
+            model_schema = cast(
+                Dict[str, Any],
+                jsonref.replace_refs(obj=model_schema, proxies=False),
+            )  # type: ignore
+            del model_schema["$defs"]
+
+        if "required" not in model_schema:
+            model_schema["required"] = []
+
+        parameters = FunctionParameters(
+            type="object",
+            properties=model_schema["properties"],
+            required=model_schema["required"],
+        )
+
+        return parameters
+
+    @classmethod
+    def verify_list_args(
+        cls,
+        args_list: List[Union[str, Dict, BaseModel]],
+    ) -> List[ToolArgsT]:
+        """Verify the list of stringify input arguments.
+
+        Args:
+            args_list: List of stringify input arguments.
+
+        Returns:
+            List[ToolArgsT]: The validated input arguments.
+        """
+        return_list = []
+        for args in args_list:
+            return_list.append(cls.verify_args(args))
+        return return_list
+
+    @classmethod
+    def verify_args(cls, args: Union[str, Dict, BaseModel]) -> BaseModel:
+        """Verify the stringify input arguments.
+
+        Args:
+            args: Stringify input arguments (string, dict, or BaseModel).
+
+        Returns:
+            BaseModel: The validated input arguments.
+
+        Raises:
+            ValueError: If JSON format is invalid or validation fails.
+        """
+        try:
+            if isinstance(args, str):
+                args_dict = json.loads(args)
+            elif isinstance(args, BaseModel):
+                args_dict = args.model_dump()
+            else:
+                args_dict = args
+        except json.JSONDecodeError as e:
+            raise ValueError(f"Invalid JSON format: {e}") from e
+
+        # Get the ArgsT type from the current class
+        args_type = get_args(cls.__orig_bases__[0])[0]
+
+        # Validate the arguments using the Pydantic model
+        try:
+            validated_args = args_type(**args_dict)
+        except ValidationError as e:
+            raise ValueError(f"Validation error: {e}") from e
+
+        return validated_args
+
+    @classmethod
+    def return_value_as_string(cls, value: ToolArgsT) -> str:
+        """Convert return value to string representation.
+
+        Args:
+            value: The value to convert to string.
+
+        Returns:
+            str: String representation of the value.
+        """
+        if isinstance(value, BaseModel):
+            dumped = value.model_dump()
+            if isinstance(dumped, dict):
+                return json.dumps(dumped)
+            return str(dumped)
+
+        return str(value)

agentscope_runtime/tools/cli/modelstudio_mcp_server.py

@@ -0,0 +1,78 @@
+# -*- coding: utf-8 -*-
+
+import os
+import sys
+import logging
+
+from mcp.server.fastmcp import FastMCP
+
+from .. import mcp_server_metas
+from ..mcp_wrapper import MCPWrapper
+
+logger = logging.getLogger(__name__)
+
+
+def main() -> None:
+    # get from args
+    server_name = None
+    for i, arg in enumerate(sys.argv):
+        logger.info(sys.argv[i])
+        logger.info(sys.argv)
+        if arg == "--server" and i + 1 < len(sys.argv):
+            server_name = sys.argv[i + 1]
+            break
+
+    # get from env
+    if not server_name:
+        server_name = os.getenv("SERVER_NAME", None)
+
+    all_server_names = set(mcp_server_metas.keys())
+    if not server_name:
+        logger.error(
+            f"Please specify the server name with --server <server_name>, "
+            f"support servers are {list(all_server_names)}",
+        )
+        sys.exit(1)
+
+    if server_name not in mcp_server_metas:
+        logger.error(
+            f"Invalid server name '{server_name}',"
+            f" Available servers: {list(all_server_names)}",
+        )
+        sys.exit(1)
+
+    port = int(os.getenv("PORT", "8080"))
+    host = os.getenv("HOST", "0.0.0.0")
+
+    # Get server metadata
+    server_meta = mcp_server_metas[server_name]
+
+    # Override server name and description if specified
+    mcp_server_name = os.getenv("OVERRIDE_NAME", server_name)
+    mcp_server_instructions = os.getenv(
+        "OVERRIDE_DESCRIPTION",
+        server_meta.instructions,
+    )
+
+    # Create an MCP server
+    mcp = FastMCP(
+        name=mcp_server_name,
+        instructions=mcp_server_instructions,
+        port=port,
+        host=host,
+    )
+
+    # Register each component as a tool
+    for component in server_meta.components:
+        MCPWrapper(mcp, component).wrap(component.name, component.description)
+        logger.info(f"Added tool: {component.name}")
+
+    logger.info("MCP server is running...")
+
+    # get mcp transport type
+    transport_type = os.getenv("TRANSPORT", "sse")
+    mcp.run(transport=transport_type)
+
+
+if __name__ == "__main__":
+    main()

agentscope_runtime/tools/generations/__init__.py

@@ -0,0 +1,75 @@
+# -*- coding: utf-8 -*-
+from .async_image_to_video import (
+    ImageToVideoSubmit,
+    ImageToVideoSubmitInput,
+    ImageToVideoFetch,
+    ImageToVideoFetchInput,
+)
+from .async_speech_to_video import (
+    SpeechToVideoSubmit,
+    SpeechToVideoSubmitInput,
+    SpeechToVideoFetch,
+    SpeechToVideoFetchInput,
+)
+from .async_text_to_video import (
+    TextToVideoSubmit,
+    TextToVideoSubmitInput,
+    TextToVideoFetch,
+    TextToVideoFetchInput,
+)
+from .image_edit import (
+    ImageEdit,
+    ImageGenInput as ImageEditInput,
+)
+from .image_generation import (
+    ImageGeneration,
+    ImageGenInput,
+)
+from .image_style_repaint import (
+    ImageStyleRepaint,
+    ImageStyleRepaintInput,
+)
+from .image_to_video import (
+    ImageToVideo,
+    ImageToVideoInput,
+)
+from .qwen_image_edit import (
+    QwenImageEdit,
+    QwenImageEditInput,
+)
+from .qwen_image_generation import (
+    QwenImageGen,
+    QwenImageGenInput,
+)
+from .qwen_text_to_speech import (
+    QwenTextToSpeech,
+    QwenTextToSpeechInput,
+)
+from .speech_to_text import (
+    SpeechToText,
+    SpeechToTextInput,
+)
+from .speech_to_video import (
+    SpeechToVideo,
+    SpeechToVideoInput,
+)
+from .async_image_to_video_wan25 import (
+    ImageToVideoWan25Submit,
+    ImageToVideoWan25SubmitInput,
+    ImageToVideoWan25Fetch,
+    ImageToVideoWan25FetchInput,
+)
+from .async_text_to_video_wan25 import (
+    TextToVideoWan25Submit,
+    TextToVideoWan25SubmitInput,
+    TextToVideoWan25Fetch,
+    TextToVideoWan25FetchInput,
+)
+from .image_edit_wan25 import (
+    ImageEditWan25,
+    ImageGenInput as ImageEditWan25Input,
+)
+from .image_generation_wan25 import (
+    ImageGenerationWan25,
+    ImageGenInput as ImageGenerationWan25Input,
+)