Files
crewAI/src/crewai/tools/tool_usage.py
2024-02-12 16:48:14 -08:00

234 lines
9.0 KiB
Python

from typing import Any, List, Union
import instructor
from langchain.prompts import PromptTemplate
from langchain_core.tools import BaseTool
from langchain_openai import ChatOpenAI
from crewai.agents.tools_handler import ToolsHandler
from crewai.telemtry import Telemetry
from crewai.tools.tool_calling import InstructorToolCalling, ToolCalling
from crewai.tools.tool_output_parser import ToolOutputParser
from crewai.utilities import I18N, Printer
class ToolUsageErrorException(Exception):
"""Exception raised for errors in the tool usage."""
def __init__(self, message: str) -> None:
self.message = message
super().__init__(self.message)
class ToolUsage:
"""
Class that represents the usage of a tool by an agent.
Attributes:
task: Task being executed.
tools_handler: Tools handler that will manage the tool usage.
tools: List of tools available for the agent.
tools_description: Description of the tools available for the agent.
tools_names: Names of the tools available for the agent.
llm: Language model to be used for the tool usage.
"""
def __init__(
self,
tools_handler: ToolsHandler,
tools: List[BaseTool],
tools_description: str,
tools_names: str,
task: Any,
llm: Any,
) -> None:
self._i18n: I18N = I18N()
self._printer: Printer = Printer()
self._telemetry: Telemetry = Telemetry()
self._run_attempts: int = 1
self._max_parsing_attempts: int = 2
self._remeber_format_after_usages: int = 3
self.tools_description = tools_description
self.tools_names = tools_names
self.tools_handler = tools_handler
self.tools = tools
self.task = task
self.llm = llm
def use(self, tool_string: str):
calling = self._tool_calling(tool_string)
if isinstance(calling, ToolUsageErrorException):
error = calling.message
self._printer.print(content=f"\n\n{error}\n", color="yellow")
return error
try:
tool = self._select_tool(calling.tool_name)
except Exception as e:
error = getattr(e, "message", str(e))
self._printer.print(content=f"\n\n{error}\n", color="yellow")
return error
return self._use(tool_string=tool_string, tool=tool, calling=calling)
def _use(
self,
tool_string: str,
tool: BaseTool,
calling: Union[ToolCalling, InstructorToolCalling],
) -> None:
if self._check_tool_repeated_usage(calling=calling):
try:
result = self._i18n.errors("task_repeated_usage").format(
tool=calling.tool_name,
tool_input=", ".join(calling.arguments.values()),
)
self._printer.print(content=f"\n\n{result}\n", color="yellow")
self._telemetry.tool_repeated_usage(
llm=self.llm, tool_name=tool.name, attempts=self._run_attempts
)
result = self._format_result(result=result)
return result
except Exception:
pass
self.tools_handler.on_tool_start(calling=calling)
result = self.tools_handler.cache.read(
tool=calling.tool_name, input=calling.arguments
)
if not result:
try:
result = tool._run(**calling.arguments)
except Exception as e:
self._run_attempts += 1
if self._run_attempts > self._max_parsing_attempts:
self._telemetry.tool_usage_error(llm=self.llm)
return ToolUsageErrorException(
self._i18n.errors("tool_usage_exception").format(error=e)
).message
return self.use(tool_string=tool_string)
self.tools_handler.on_tool_end(calling=calling, output=result)
self._printer.print(content=f"\n\n{result}\n", color="yellow")
self._telemetry.tool_usage(
llm=self.llm, tool_name=tool.name, attempts=self._run_attempts
)
result = self._format_result(result=result)
return result
def _format_result(self, result: Any) -> None:
self.task.used_tools += 1
if self._should_remember_format():
result = self._remember_format(result=result)
return result
def _should_remember_format(self) -> None:
return self.task.used_tools % self._remeber_format_after_usages == 0
def _remember_format(self, result: str) -> None:
result = str(result)
result += "\n\n" + self._i18n.slice("tools").format(
tools=self.tools_description, tool_names=self.tools_names
)
return result
def _check_tool_repeated_usage(
self, calling: Union[ToolCalling, InstructorToolCalling]
) -> None:
if last_tool_usage := self.tools_handler.last_used_tool:
return calling == last_tool_usage
def _select_tool(self, tool_name: str) -> BaseTool:
for tool in self.tools:
if tool.name.lower().strip() == tool_name.lower().strip():
return tool
raise Exception(f"Tool '{tool_name}' not found.")
def _render(self) -> str:
"""Render the tool name and description in plain text."""
descriptions = []
for tool in self.tools:
args = {
k: {k2: v2 for k2, v2 in v.items() if k2 in ["description", "type"]}
for k, v in tool.args.items()
}
descriptions.append(
"\n".join(
[
f"Tool Name: {tool.name.lower()}",
f"Tool Description: {tool.description}",
f"Tool Arguments: {args}",
]
)
)
return "\n--\n".join(descriptions)
def _tool_calling(
self, tool_string: str
) -> Union[ToolCalling, InstructorToolCalling]:
try:
tool_string = tool_string.replace(
"Thought: Do I need to use a tool? Yes", ""
)
tool_string = tool_string.replace("Action:", "Tool Name:")
tool_string = tool_string.replace("Action Input:", "Tool Arguments:")
if (isinstance(self.llm, ChatOpenAI)) and (
self.llm.openai_api_base == None
):
client = instructor.patch(
self.llm.client._client,
mode=instructor.Mode.FUNCTIONS,
)
calling = client.chat.completions.create(
model=self.llm.model_name,
messages=[
{
"role": "system",
"content": """
The schema should have the following structure, only two key:
- tool_name: str
- arguments: dict (with all arguments being passed)
Example:
{"tool_name": "tool_name", "arguments": {"arg_name1": "value", "arg_name2": 2}}
""",
},
{
"role": "user",
"content": f"Tools available:\n\n{self._render()}\n\nReturn a valid schema for the tool, use this text to inform a valid ouput schema:\n{tool_string}```",
},
],
response_model=InstructorToolCalling,
)
else:
parser = ToolOutputParser(pydantic_object=ToolCalling)
prompt = PromptTemplate(
template="Tools available:\n\n{available_tools}\n\nReturn a valid schema for the tool, use this text to inform a valid ouput schema:\n{tool_string}\n\n{format_instructions}\n```",
input_variables=["tool_string"],
partial_variables={
"available_tools": self._render(),
"format_instructions": """
The schema should have the following structure, only two key:
- tool_name: str
- arguments: dict (with all arguments being passed)
Example:
{"tool_name": "tool_name", "arguments": {"arg_name1": "value", "arg_name2": 2}}
""",
},
)
chain = prompt | self.llm | parser
calling = chain.invoke({"tool_string": tool_string})
except Exception:
self._run_attempts += 1
if self._run_attempts > self._max_parsing_attempts:
self._telemetry.tool_usage_error(llm=self.llm)
return ToolUsageErrorException(self._i18n.errors("tool_usage_error"))
return self._tool_calling(tool_string)
return calling