|
| 1 | +# SPDX-FileCopyrightText: 2022-present deepset GmbH <[email protected]> |
| 2 | +# |
| 3 | +# SPDX-License-Identifier: Apache-2.0 |
| 4 | + |
| 5 | +import inspect |
| 6 | +from dataclasses import asdict, dataclass |
| 7 | +from typing import Any, Callable, Dict, Optional |
| 8 | + |
| 9 | +from pydantic import create_model |
| 10 | + |
| 11 | +from haystack.lazy_imports import LazyImport |
| 12 | +from haystack.utils import deserialize_callable, serialize_callable |
| 13 | + |
| 14 | +with LazyImport(message="Run 'pip install jsonschema'") as jsonschema_import: |
| 15 | + from jsonschema import Draft202012Validator |
| 16 | + from jsonschema.exceptions import SchemaError |
| 17 | + |
| 18 | + |
| 19 | +class ToolInvocationError(Exception): |
| 20 | + """ |
| 21 | + Exception raised when a Tool invocation fails. |
| 22 | + """ |
| 23 | + |
| 24 | + pass |
| 25 | + |
| 26 | + |
| 27 | +class SchemaGenerationError(Exception): |
| 28 | + """ |
| 29 | + Exception raised when automatic schema generation fails. |
| 30 | + """ |
| 31 | + |
| 32 | + pass |
| 33 | + |
| 34 | + |
| 35 | +@dataclass |
| 36 | +class Tool: |
| 37 | + """ |
| 38 | + Data class representing a Tool that Language Models can prepare a call for. |
| 39 | +
|
| 40 | + Accurate definitions of the textual attributes such as `name` and `description` |
| 41 | + are important for the Language Model to correctly prepare the call. |
| 42 | +
|
| 43 | + :param name: |
| 44 | + Name of the Tool. |
| 45 | + :param description: |
| 46 | + Description of the Tool. |
| 47 | + :param parameters: |
| 48 | + A JSON schema defining the parameters expected by the Tool. |
| 49 | + :param function: |
| 50 | + The function that will be invoked when the Tool is called. |
| 51 | + """ |
| 52 | + |
| 53 | + name: str |
| 54 | + description: str |
| 55 | + parameters: Dict[str, Any] |
| 56 | + function: Callable |
| 57 | + |
| 58 | + def __post_init__(self): |
| 59 | + jsonschema_import.check() |
| 60 | + # Check that the parameters define a valid JSON schema |
| 61 | + try: |
| 62 | + Draft202012Validator.check_schema(self.parameters) |
| 63 | + except SchemaError as e: |
| 64 | + raise ValueError("The provided parameters do not define a valid JSON schema") from e |
| 65 | + |
| 66 | + @property |
| 67 | + def tool_spec(self) -> Dict[str, Any]: |
| 68 | + """ |
| 69 | + Return the Tool specification to be used by the Language Model. |
| 70 | + """ |
| 71 | + return {"name": self.name, "description": self.description, "parameters": self.parameters} |
| 72 | + |
| 73 | + def invoke(self, **kwargs) -> Any: |
| 74 | + """ |
| 75 | + Invoke the Tool with the provided keyword arguments. |
| 76 | + """ |
| 77 | + |
| 78 | + try: |
| 79 | + result = self.function(**kwargs) |
| 80 | + except Exception as e: |
| 81 | + raise ToolInvocationError(f"Failed to invoke Tool `{self.name}` with parameters {kwargs}") from e |
| 82 | + return result |
| 83 | + |
| 84 | + def to_dict(self) -> Dict[str, Any]: |
| 85 | + """ |
| 86 | + Serializes the Tool to a dictionary. |
| 87 | +
|
| 88 | + :returns: |
| 89 | + Dictionary with serialized data. |
| 90 | + """ |
| 91 | + |
| 92 | + serialized = asdict(self) |
| 93 | + serialized["function"] = serialize_callable(self.function) |
| 94 | + return serialized |
| 95 | + |
| 96 | + @classmethod |
| 97 | + def from_dict(cls, data: Dict[str, Any]) -> "Tool": |
| 98 | + """ |
| 99 | + Deserializes the Tool from a dictionary. |
| 100 | +
|
| 101 | + :param data: |
| 102 | + Dictionary to deserialize from. |
| 103 | + :returns: |
| 104 | + Deserialized Tool. |
| 105 | + """ |
| 106 | + data["function"] = deserialize_callable(data["function"]) |
| 107 | + return cls(**data) |
| 108 | + |
| 109 | + @classmethod |
| 110 | + def from_function(cls, function: Callable, name: Optional[str] = None, description: Optional[str] = None) -> "Tool": |
| 111 | + """ |
| 112 | + Create a Tool instance from a function. |
| 113 | +
|
| 114 | + ### Usage example |
| 115 | +
|
| 116 | + ```python |
| 117 | + from typing import Annotated, Literal |
| 118 | + from haystack.dataclasses import Tool |
| 119 | +
|
| 120 | + def get_weather( |
| 121 | + city: Annotated[str, "the city for which to get the weather"] = "Munich", |
| 122 | + unit: Annotated[Literal["Celsius", "Fahrenheit"], "the unit for the temperature"] = "Celsius"): |
| 123 | + '''A simple function to get the current weather for a location.''' |
| 124 | + return f"Weather report for {city}: 20 {unit}, sunny" |
| 125 | +
|
| 126 | + tool = Tool.from_function(get_weather) |
| 127 | +
|
| 128 | + print(tool) |
| 129 | + >>> Tool(name='get_weather', description='A simple function to get the current weather for a location.', |
| 130 | + >>> parameters={ |
| 131 | + >>> 'type': 'object', |
| 132 | + >>> 'properties': { |
| 133 | + >>> 'city': {'type': 'string', 'description': 'the city for which to get the weather', 'default': 'Munich'}, |
| 134 | + >>> 'unit': { |
| 135 | + >>> 'type': 'string', |
| 136 | + >>> 'enum': ['Celsius', 'Fahrenheit'], |
| 137 | + >>> 'description': 'the unit for the temperature', |
| 138 | + >>> 'default': 'Celsius', |
| 139 | + >>> }, |
| 140 | + >>> } |
| 141 | + >>> }, |
| 142 | + >>> function=<function get_weather at 0x7f7b3a8a9b80>) |
| 143 | + ``` |
| 144 | +
|
| 145 | + :param function: |
| 146 | + The function to be converted into a Tool. |
| 147 | + The function must include type hints for all parameters. |
| 148 | + If a parameter is annotated using `typing.Annotated`, its metadata will be used as parameter description. |
| 149 | + :param name: |
| 150 | + The name of the Tool. If not provided, the name of the function will be used. |
| 151 | + :param description: |
| 152 | + The description of the Tool. If not provided, the docstring of the function will be used. |
| 153 | + To intentionally leave the description empty, pass an empty string. |
| 154 | +
|
| 155 | + :returns: |
| 156 | + The Tool created from the function. |
| 157 | +
|
| 158 | + :raises ValueError: |
| 159 | + If any parameter of the function lacks a type hint. |
| 160 | + :raises SchemaGenerationError: |
| 161 | + If there is an error generating the JSON schema for the Tool. |
| 162 | + """ |
| 163 | + |
| 164 | + tool_description = description if description is not None else (function.__doc__ or "") |
| 165 | + |
| 166 | + signature = inspect.signature(function) |
| 167 | + |
| 168 | + # collect fields (types and defaults) and descriptions from function parameters |
| 169 | + fields: Dict[str, Any] = {} |
| 170 | + descriptions = {} |
| 171 | + |
| 172 | + for param_name, param in signature.parameters.items(): |
| 173 | + if param.annotation is param.empty: |
| 174 | + raise ValueError(f"Function '{function.__name__}': parameter '{param_name}' does not have a type hint.") |
| 175 | + |
| 176 | + # if the parameter has not a default value, Pydantic requires an Ellipsis (...) |
| 177 | + # to explicitly indicate that the parameter is required |
| 178 | + default = param.default if param.default is not param.empty else ... |
| 179 | + fields[param_name] = (param.annotation, default) |
| 180 | + |
| 181 | + if hasattr(param.annotation, "__metadata__"): |
| 182 | + descriptions[param_name] = param.annotation.__metadata__[0] |
| 183 | + |
| 184 | + # create Pydantic model and generate JSON schema |
| 185 | + try: |
| 186 | + model = create_model(function.__name__, **fields) |
| 187 | + schema = model.model_json_schema() |
| 188 | + except Exception as e: |
| 189 | + raise SchemaGenerationError(f"Failed to create JSON schema for function '{function.__name__}'") from e |
| 190 | + |
| 191 | + # we don't want to include title keywords in the schema, as they contain redundant information |
| 192 | + # there is no programmatic way to prevent Pydantic from adding them, so we remove them later |
| 193 | + # see https://github.com/pydantic/pydantic/discussions/8504 |
| 194 | + _remove_title_from_schema(schema) |
| 195 | + |
| 196 | + # add parameters descriptions to the schema |
| 197 | + for param_name, param_description in descriptions.items(): |
| 198 | + if param_name in schema["properties"]: |
| 199 | + schema["properties"][param_name]["description"] = param_description |
| 200 | + |
| 201 | + return Tool(name=name or function.__name__, description=tool_description, parameters=schema, function=function) |
| 202 | + |
| 203 | + |
| 204 | +def _remove_title_from_schema(schema: Dict[str, Any]): |
| 205 | + """ |
| 206 | + Remove the 'title' keyword from JSON schema and contained property schemas. |
| 207 | +
|
| 208 | + :param schema: |
| 209 | + The JSON schema to remove the 'title' keyword from. |
| 210 | + """ |
| 211 | + schema.pop("title", None) |
| 212 | + |
| 213 | + for property_schema in schema["properties"].values(): |
| 214 | + for key in list(property_schema.keys()): |
| 215 | + if key == "title": |
| 216 | + del property_schema[key] |
| 217 | + |
| 218 | + |
| 219 | +def deserialize_tools_inplace(data: Dict[str, Any], key: str = "tools"): |
| 220 | + """ |
| 221 | + Deserialize Tools in a dictionary inplace. |
| 222 | +
|
| 223 | + :param data: |
| 224 | + The dictionary with the serialized data. |
| 225 | + :param key: |
| 226 | + The key in the dictionary where the Tools are stored. |
| 227 | + """ |
| 228 | + if key in data: |
| 229 | + serialized_tools = data[key] |
| 230 | + |
| 231 | + if serialized_tools is None: |
| 232 | + return |
| 233 | + |
| 234 | + if not isinstance(serialized_tools, list): |
| 235 | + raise TypeError(f"The value of '{key}' is not a list") |
| 236 | + |
| 237 | + deserialized_tools = [] |
| 238 | + for tool in serialized_tools: |
| 239 | + if not isinstance(tool, dict): |
| 240 | + raise TypeError(f"Serialized tool '{tool}' is not a dictionary") |
| 241 | + deserialized_tools.append(Tool.from_dict(tool)) |
| 242 | + |
| 243 | + data[key] = deserialized_tools |
0 commit comments