andre/crewAI
1
0
Fork 0
mirror of https://github.com/crewAIInc/crewAI.git synced 2026-01-08 15:48:29 +00:00
Code Issues Actions 4 Packages Projects Releases Wiki Activity

Revamping tool usage

Browse Source
This commit is contained in:
João Moura
2024-02-10 10:36:34 -08:00
parent 58bb181b48
commit 5a102251cf
22 changed files with 1233 additions and 584 deletions
Download Patch File Download Diff File Expand all files Collapse all files

23
src/crewai/agent.py
View File

@@ -3,9 +3,9 @@ from typing import Any, List, Optional
from langchain.agents.agent import RunnableAgent
from langchain.agents.format_scratchpad import format_log_to_str
from langchain.agents.output_parsers import ReActSingleInputOutputParser
from langchain.memory import ConversationSummaryMemory
from langchain.tools.render import render_text_description
from langchain_core.runnables.config import RunnableConfig
from langchain_openai import ChatOpenAI
from pydantic import (
UUID4,
@@ -19,12 +19,7 @@ from pydantic import (
)
from pydantic_core import PydanticCustomError
from crewai.agents import (
CacheHandler,
CrewAgentExecutor,
CrewAgentOutputParser,
ToolsHandler,
)
from crewai.agents import CacheHandler, CrewAgentExecutor, ToolsHandler
from crewai.utilities import I18N, Logger, Prompts, RPMController
@@ -158,8 +153,7 @@ class Agent(BaseModel):
"input": task,
"tool_names": self.__tools_names(tools),
"tools": render_text_description(tools),
},
RunnableConfig(callbacks=[self.tools_handler]),
}
)["output"]
if self.max_rpm:
@@ -200,12 +194,14 @@ class Agent(BaseModel):
"agent_scratchpad": lambda x: format_log_to_str(x["intermediate_steps"]),
}
executor_args = {
"llm": self.llm,
"i18n": self.i18n,
"tools": self.tools,
"verbose": self.verbose,
"handle_parsing_errors": True,
"max_iterations": self.max_iter,
"step_callback": self.step_callback,
"tools_handler": self.tools_handler,
}
if self._rpm_controller:
@@ -231,14 +227,7 @@ class Agent(BaseModel):
bind = self.llm.bind(stop=[self.i18n.slice("observation")])
inner_agent = (
agent_args
| execution_prompt
| bind
| CrewAgentOutputParser(
tools_handler=self.tools_handler,
cache=self.cache_handler,
i18n=self.i18n,
)
agent_args | execution_prompt | bind | ReActSingleInputOutputParser()
)
self.agent_executor = CrewAgentExecutor(
agent=RunnableAgent(runnable=inner_agent), **executor_args

1
src/crewai/agents/__init__.py
View File

@@ -1,4 +1,3 @@
from .cache.cache_handler import CacheHandler
from .executor import CrewAgentExecutor
from .output_parser import CrewAgentOutputParser
from .tools_handler import ToolsHandler

1
src/crewai/agents/cache/__init__.py vendored
View File

@@ -1,2 +1 @@
from .cache_handler import CacheHandler
from .cache_hit import CacheHit

2
src/crewai/agents/cache/cache_handler.py vendored
View File

@@ -10,9 +10,7 @@ class CacheHandler:
self._cache = {}
def add(self, tool, input, output):
input = input.strip()
self._cache[f"{tool}-{input}"] = output
def read(self, tool, input) -> Optional[str]:
input = input.strip()
return self._cache.get(f"{tool}-{input}")

18
src/crewai/agents/cache/cache_hit.py vendored
View File

@@ -1,18 +0,0 @@
from typing import Any
from pydantic import BaseModel, Field
from .cache_handler import CacheHandler
class CacheHit(BaseModel):
"""Cache Hit Object."""
class Config:
arbitrary_types_allowed = True
# Making it Any instead of AgentAction to avoind
# pydantic v1 vs v2 incompatibility, langchain should
# soon be updated to pydantic v2
action: Any = Field(description="Action taken")
cache: CacheHandler = Field(description="Cache Handler for the tool")

30
src/crewai/agents/exceptions.py
View File

@@ -1,30 +0,0 @@
from langchain_core.exceptions import OutputParserException
from crewai.utilities import I18N
class TaskRepeatedUsageException(OutputParserException):
"""Exception raised when a task is used twice in a roll."""
i18n: I18N = I18N()
error: str = "TaskRepeatedUsageException"
message: str
def __init__(self, i18n: I18N, tool: str, tool_input: str, text: str):
self.i18n = i18n
self.text = text
self.tool = tool
self.tool_input = tool_input
self.message = self.i18n.errors("task_repeated_usage").format(
tool=tool, tool_input=tool_input
)
super().__init__(
error=self.error,
observation=self.message,
send_to_llm=True,
llm_output=self.text,
)
def __str__(self):
return self.message

47
src/crewai/agents/executor.py
View File

@@ -10,16 +10,19 @@ from langchain_core.exceptions import OutputParserException
from langchain_core.pydantic_v1 import root_validator
from langchain_core.tools import BaseTool
from langchain_core.utils.input import get_color_mapping
from pydantic import InstanceOf
from crewai.agents.cache.cache_hit import CacheHit
from crewai.tools.cache_tools import CacheTools
from crewai.agents.tools_handler import ToolsHandler
from crewai.tools.tool_usage import ToolUsage
from crewai.utilities import I18N
class CrewAgentExecutor(AgentExecutor):
i18n: I18N = I18N()
llm: Any = None
iterations: int = 0
request_within_rpm_limit: Any = None
tools_handler: InstanceOf[ToolsHandler] = None
max_iterations: Optional[int] = 15
force_answer_max_iterations: Optional[int] = None
step_callback: Optional[Any] = None
@@ -32,11 +35,6 @@ class CrewAgentExecutor(AgentExecutor):
def _should_force_answer(self) -> bool:
return True if self.iterations == self.force_answer_max_iterations else False
def _force_answer(self, output: AgentAction):
return AgentStep(
action=output, observation=self.i18n.errors("force_final_answer")
)
def _call(
self,
inputs: Dict[str, str],
@@ -110,16 +108,17 @@ class CrewAgentExecutor(AgentExecutor):
callbacks=run_manager.get_child() if run_manager else None,
**inputs,
)
if self._should_force_answer():
if isinstance(output, AgentAction) or isinstance(output, AgentFinish):
output = output
elif isinstance(output, CacheHit):
output = output.action
else:
raise ValueError(
f"Unexpected output type from agent: {type(output)}"
)
yield self._force_answer(output)
yield AgentStep(
action=output, observation=self.i18n.errors("force_final_answer")
)
return
except OutputParserException as e:
@@ -160,7 +159,9 @@ class CrewAgentExecutor(AgentExecutor):
)
if self._should_force_answer():
yield self._force_answer(output)
yield AgentStep(
action=output, observation=self.i18n.errors("force_final_answer")
)
return
yield AgentStep(action=output, observation=observation)
@@ -171,17 +172,6 @@ class CrewAgentExecutor(AgentExecutor):
yield output
return
# Override tool usage to use CacheTools
if isinstance(output, CacheHit):
cache = output.cache
action = output.action
tool = CacheTools(cache_handler=cache).tool()
output = action.copy()
output.tool_input = f"tool:{action.tool}|input:{action.tool_input}"
output.tool = tool.name
name_to_tool_map[tool.name] = tool
color_mapping[tool.name] = color_mapping[action.tool]
actions: List[AgentAction]
actions = [output] if isinstance(output, AgentAction) else output
yield from actions
@@ -192,18 +182,13 @@ class CrewAgentExecutor(AgentExecutor):
if agent_action.tool in name_to_tool_map:
tool = name_to_tool_map[agent_action.tool]
return_direct = tool.return_direct
color = color_mapping[agent_action.tool]
color_mapping[agent_action.tool]
tool_run_kwargs = self.agent.tool_run_logging_kwargs()
if return_direct:
tool_run_kwargs["llm_prefix"] = ""
# We then call the tool on the tool input to get an observation
observation = tool.run(
agent_action.tool_input,
verbose=self.verbose,
color=color,
callbacks=run_manager.get_child() if run_manager else None,
**tool_run_kwargs,
)
observation = ToolUsage(
tools_handler=self.tools_handler, tools=self.tools, llm=self.llm
).use(agent_action.log)
else:
tool_run_kwargs = self.agent.tool_run_logging_kwargs()
observation = InvalidTool().run(

79
src/crewai/agents/output_parser.py
View File

@@ -1,79 +0,0 @@
import re
from typing import Union
from langchain.agents.output_parsers import ReActSingleInputOutputParser
from langchain_core.agents import AgentAction, AgentFinish
from crewai.agents.cache import CacheHandler, CacheHit
from crewai.agents.exceptions import TaskRepeatedUsageException
from crewai.agents.tools_handler import ToolsHandler
from crewai.utilities import I18N
FINAL_ANSWER_ACTION = "Final Answer:"
FINAL_ANSWER_AND_PARSABLE_ACTION_ERROR_MESSAGE = (
"Parsing LLM output produced both a final answer and a parse-able action:"
)
class CrewAgentOutputParser(ReActSingleInputOutputParser):
"""Parses ReAct-style LLM calls that have a single tool input.
Expects output to be in one of two formats.
If the output signals that an action should be taken,
should be in the below format. This will result in an AgentAction
being returned.
```
Thought: agent thought here
Action: search
Action Input: what is the temperature in SF?
```
If the output signals that a final answer should be given,
should be in the below format. This will result in an AgentFinish
being returned.
```
Thought: agent thought here
Final Answer: The temperature is 100 degrees
```
It also prevents tools from being reused in a roll.
"""
class Config:
arbitrary_types_allowed = True
tools_handler: ToolsHandler
cache: CacheHandler
i18n: I18N
def parse(self, text: str) -> Union[AgentAction, AgentFinish, CacheHit]:
regex = (
r"Action\s*\d*\s*:[\s]*(.*?)[\s]*Action\s*\d*\s*Input\s*\d*\s*:[\s]*(.*)"
)
if action_match := re.search(regex, text, re.DOTALL):
action = action_match.group(1).strip()
action_input = action_match.group(2)
tool_input = action_input.strip(" ")
tool_input = tool_input.strip('"')
if last_tool_usage := self.tools_handler.last_used_tool:
usage = {
"tool": action,
"input": tool_input,
}
if usage == last_tool_usage:
raise TaskRepeatedUsageException(
text=text,
tool=action,
tool_input=tool_input,
i18n=self.i18n,
)
if self.cache.read(action, tool_input):
action = AgentAction(action, tool_input, text)
return CacheHit(action=action, cache=self.cache)
return super().parse(text)

44
src/crewai/agents/tools_handler.py
View File

@@ -1,44 +1,30 @@
from typing import Any, Dict
from langchain.callbacks.base import BaseCallbackHandler
from typing import Any
from ..tools.cache_tools import CacheTools
from ..tools.tool_calling import ToolCalling
from .cache.cache_handler import CacheHandler
class ToolsHandler(BaseCallbackHandler):
class ToolsHandler:
"""Callback handler for tool usage."""
last_used_tool: Dict[str, Any] = {}
last_used_tool: ToolCalling = {}
cache: CacheHandler
def __init__(self, cache: CacheHandler, **kwargs: Any):
def __init__(self, cache: CacheHandler):
"""Initialize the callback handler."""
self.cache = cache
super().__init__(**kwargs)
self.last_used_tool = {}
def on_tool_start(
self, serialized: Dict[str, Any], input_str: str, **kwargs: Any
) -> Any:
def on_tool_start(self, calling: ToolCalling) -> Any:
"""Run when tool starts running."""
name = serialized.get("name")
if name not in ["invalid_tool", "_Exception"]:
tools_usage = {
"tool": name,
"input": input_str,
}
self.last_used_tool = tools_usage
self.last_used_tool = calling
def on_tool_end(self, output: str, **kwargs: Any) -> Any:
def on_tool_end(self, calling: ToolCalling, output: str) -> Any:
"""Run when tool ends running."""
if (
"is not a valid tool" not in output
and "Invalid or incomplete response" not in output
and "Invalid Format" not in output
):
if self.last_used_tool["tool"] != CacheTools().name:
self.cache.add(
tool=self.last_used_tool["tool"],
input=self.last_used_tool["input"],
output=output,
)
if self.last_used_tool.function_name != CacheTools().name:
self.cache.add(
tool=calling.function_name,
input=calling.arguments,
output=output,
)

1
src/crewai/crew.py
View File

@@ -107,6 +107,7 @@ class Crew(BaseModel):
self._logger = Logger(self.verbose)
self._rpm_controller = RPMController(max_rpm=self.max_rpm, logger=self._logger)
self._telemetry = Telemetry()
self._telemetry.set_tracer()
self._telemetry.crew_creation(self)
return self

39
src/crewai/telemtry/telemetry.py
View File

@@ -2,6 +2,7 @@ import json
import os
import platform
import socket
from typing import Any
import pkg_resources
from opentelemetry import trace
@@ -42,16 +43,18 @@ class Telemetry:
try:
telemetry_endpoint = "http://telemetry.crewai.com:4318"
self.resource = Resource(attributes={SERVICE_NAME: "crewAI-telemetry"})
provider = TracerProvider(resource=self.resource)
self.provider = TracerProvider(resource=self.resource)
processor = BatchSpanProcessor(
OTLPSpanExporter(endpoint=f"{telemetry_endpoint}/v1/traces")
)
provider.add_span_processor(processor)
trace.set_tracer_provider(provider)
self.provider.add_span_processor(processor)
self.ready = True
except Exception:
pass
def set_tracer(self):
trace.set_tracer_provider(self.provider)
def crew_creation(self, crew):
"""Records the creation of a crew."""
if self.ready:
@@ -116,6 +119,36 @@ class Telemetry:
except Exception:
pass
def tool_usage(self, llm: Any, tool_name: str, attempts: int):
"""Records the usage of a tool by an agent."""
if self.ready:
try:
tracer = trace.get_tracer("crewai.telemetry")
span = tracer.start_span("Tool Usage")
self._add_attribute(span, "tool_name", tool_name)
self._add_attribute(span, "attempts", attempts)
self._add_attribute(
span, "llm", json.dumps(self._safe_llm_attributes(llm))
)
span.set_status(Status(StatusCode.OK))
span.end()
except Exception:
pass
def tool_usage_error(self, llm: Any):
"""Records the usage of a tool by an agent."""
if self.ready:
try:
tracer = trace.get_tracer("crewai.telemetry")
span = tracer.start_span("Tool Usage Error")
self._add_attribute(
span, "llm", json.dumps(self._safe_llm_attributes(llm))
)
span.set_status(Status(StatusCode.OK))
span.end()
except Exception:
pass
def crew_execution_span(self, crew):
"""Records the complete execution of a crew.
This is only collected if the user has opted-in to share the crew.

24
src/crewai/tools/agent_tools.py
View File

@@ -1,6 +1,6 @@
from typing import List
from langchain.tools import Tool
from langchain.tools import StructuredTool
from pydantic import BaseModel, Field
from crewai.agent import Agent
@@ -15,14 +15,14 @@ class AgentTools(BaseModel):
def tools(self):
return [
Tool.from_function(
StructuredTool.from_function(
func=self.delegate_work,
name="Delegate work to co-worker",
description=self.i18n.tools("delegate_work").format(
coworkers=", ".join([agent.role for agent in self.agents])
),
),
Tool.from_function(
StructuredTool.from_function(
func=self.ask_question,
name="Ask question to co-worker",
description=self.i18n.tools("ask_question").format(
@@ -31,24 +31,16 @@ class AgentTools(BaseModel):
),
]
def delegate_work(self, command):
def delegate_work(self, coworker: str, task: str, context: str):
"""Useful to delegate a specific task to a coworker."""
return self._execute(command)
return self._execute(coworker, task, context)
def ask_question(self, command):
def ask_question(self, coworker: str, question: str, context: str):
"""Useful to ask a question, opinion or take from a coworker."""
return self._execute(command)
return self._execute(coworker, question, context)
def _execute(self, command):
def _execute(self, agent, task, context):
"""Execute the command."""
try:
agent, task, context = command.split("|")
except ValueError:
return self.i18n.errors("agent_tool_missing_param")
if not agent or not task or not context:
return self.i18n.errors("agent_tool_missing_param")
agent = [
available_agent
for available_agent in self.agents

4
src/crewai/tools/cache_tools.py
View File

@@ -1,4 +1,4 @@
from langchain.tools import Tool
from langchain.tools import StructuredTool
from pydantic import BaseModel, ConfigDict, Field
from crewai.agents.cache import CacheHandler
@@ -15,7 +15,7 @@ class CacheTools(BaseModel):
)
def tool(self):
return Tool.from_function(
return StructuredTool.from_function(
func=self.hit_cache,
name=self.name,
description="Reads directly from the cache",

12
src/crewai/tools/tool_calling.py Normal file
View File

@@ -0,0 +1,12 @@
from typing import Any, Dict
from pydantic.v1 import BaseModel, Field
class ToolCalling(BaseModel):
function_name: str = Field(
..., description="The name of the function to be called."
)
arguments: Dict[str, Any] = Field(
..., description="A dictinary of arguments to be passed to the function."
)

112
src/crewai/tools/tool_usage.py Normal file
View File

@@ -0,0 +1,112 @@
from typing import Any, List
from langchain.output_parsers import PydanticOutputParser
from langchain.prompts import PromptTemplate
from langchain_core.tools import BaseTool
from crewai.agents.tools_handler import ToolsHandler
from crewai.telemtry import Telemetry
from crewai.tools.tool_calling import ToolCalling
from crewai.utilities import I18N, Printer
class ToolUsage:
"""
Class that represents the usage of a tool by an agent.
Attributes:
tools_handler: Tools handler that will manage the tool usage.
tools: List of tools available for the agent.
llm: Language model to be used for the tool usage.
"""
def __init__(
self, tools_handler: ToolsHandler, tools: List[BaseTool], 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 = 3
self.tools_handler = tools_handler
self.tools = tools
self.llm = llm
def use(self, tool_string: str):
calling = self._tool_calling(tool_string)
tool = self._select_tool(calling.function_name)
return self._use(tool=tool, calling=calling)
def _use(self, tool: BaseTool, calling: ToolCalling) -> None:
if self._check_tool_repeated_usage(calling=calling):
result = self._i18n.errors("task_repeated_usage").format(
tool=calling.function_name, tool_input=calling.arguments
)
else:
self.tools_handler.on_tool_start(calling=calling)
result = self.tools_handler.cache.read(
tool=calling.function_name, input=calling.arguments
)
if not result:
result = tool._run(**calling.arguments)
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
)
return result
def _check_tool_repeated_usage(self, calling: ToolCalling) -> 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 == tool_name:
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"Funtion Name: {tool.name}",
f"Funtion attributes: {args}",
f"Description: {tool.description}",
]
)
)
return "\n--\n".join(descriptions)
def _tool_calling(self, tool_string: str) -> ToolCalling:
try:
parser = PydanticOutputParser(pydantic_object=ToolCalling)
prompt = PromptTemplate(
template="Return a valid schema for the one tool you must use with its arguments and values.\n\nTools available:\n\n{available_tools}\n\nUse 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": parser.get_format_instructions(),
},
)
chain = prompt | self.llm | parser
calling = chain.invoke({"tool_string": tool_string})
except Exception as e:
self._run_attempts += 1
if self._run_attempts > self._max_parsing_attempts:
self._telemetry.tool_usage_error(llm=self.llm)
raise e
return self._tool_calling(tool_string)
return calling

2
src/crewai/translations/el.json
View File

@@ -16,7 +16,7 @@
"errors": {
"force_final_answer": "Στην πραγματικότητα, χρησιμοποίησα πάρα πολλά εργαλεία, οπότε θα σταματήσω τώρα και θα σας δώσω την απόλυτη ΚΑΛΥΤΕΡΗ τελική μου απάντηση ΤΩΡΑ, χρησιμοποιώντας την αναμενόμενη μορφή: ```\nΣκέφτηκα: Χρειάζεται να χρησιμοποιήσω ένα εργαλείο; Όχι\nΤελική απάντηση: [η απάντησή σας εδώ]```",
"agent_tool_unexsiting_coworker": "\nΣφάλμα κατά την εκτέλεση του εργαλείου. Ο συνάδελφος που αναφέρεται στο Ενέργεια προς εισαγωγή δεν βρέθηκε, πρέπει να είναι μία από τις ακόλουθες επιλογές: {coworkers}.\n",
"task_repeated_usage": "Μόλις χρησιμοποίησα το {tool} εργαλείο με είσοδο {tool_input}. Άρα ξέρω ήδη το αποτέλεσμα αυτού και δεν χρειάζεται να το χρησιμοποιήσω τώρα.\n"
"task_repeated_usage": "Μόλις χρησιμοποίησα το εργαλείο {tool} με είσοδο {tool_input}. Άρα ξέρω ήδη το αποτέλεσμα αυτού και δεν χρειάζεται να το χρησιμοποιήσω ξανά τώρα.\n"
},
"tools": {
"delegate_work": "Αναθέστε μια συγκεκριμένη εργασία σε έναν από τους παρακάτω συναδέλφους: {coworkers}. Η είσοδος σε αυτό το εργαλείο θα πρέπει να είναι ο ρόλος του συναδέλφου, η εργασία που θέλετε να κάνει και ΟΛΟ το απαραίτητο πλαίσιο για την εκτέλεση της εργασίας, δεν γνωρίζουν τίποτα για την εργασία, επομένως μοιραστείτε απολύτως όλα όσα γνωρίζετε, μην αναφέρετε πράγματα, αλλά αντί να τους εξηγήσεις.",

2
src/crewai/translations/en.json
View File

@@ -16,7 +16,7 @@
"errors": {
"force_final_answer": "Actually, I used too many tools, so I'll stop now and give you my absolute BEST Final answer NOW, using the expected format: ```\nThought: Do I need to use a tool? No\nFinal Answer: [your response here]```",
"agent_tool_unexsiting_coworker": "\nError executing tool. Co-worker mentioned on the Action Input not found, it must to be one of the following options: {coworkers}.\n",
"task_repeated_usage": "I just used the {tool} tool with input {tool_input}. So I already know the result of that and don't need to use it now.\n"
"task_repeated_usage": "I just used the {tool} tool with input {tool_input}. So I already know the result of that and don't need to use it again now.\n"
},
"tools": {
"delegate_work": "Delegate a specific task to one of the following co-workers: {coworkers}. The input to this tool should be the role of the coworker, the task you want them to do, and ALL necessary context to exectue the task, they know nothing about the task, so share absolute everything you know, don't reference things but instead explain them.",

118
tests/agent_test.py
View File

@@ -9,6 +9,7 @@ from langchain_openai import ChatOpenAI
from crewai import Agent, Crew, Task
from crewai.agents.cache import CacheHandler
from crewai.agents.executor import CrewAgentExecutor
from crewai.tools.tool_calling import ToolCalling
from crewai.utilities import RPMController
@@ -85,13 +86,9 @@ def test_agent_execution():
@pytest.mark.vcr(filter_headers=["authorization"])
def test_agent_execution_with_tools():
@tool
def multiplier(numbers) -> float:
"""Useful for when you need to multiply two numbers together.
The input to this tool should be a comma separated list of numbers of
length two, representing the two numbers you want to multiply together.
For example, `1,2` would be the input if you wanted to multiply 1 by 2."""
a, b = numbers.split(",")
return int(a) * int(b)
def multiplier(first_number: int, second_number: int) -> float:
"""Useful for when you need to multiply two numbers together."""
return first_number * second_number
agent = Agent(
role="test role",
@@ -102,19 +99,15 @@ def test_agent_execution_with_tools():
)
output = agent.execute_task("What is 3 times 4")
assert output == "12"
assert output == "3 times 4 is 12."
@pytest.mark.vcr(filter_headers=["authorization"])
def test_logging_tool_usage():
@tool
def multiplier(numbers) -> float:
"""Useful for when you need to multiply two numbers together.
The input to this tool should be a comma separated list of numbers of
length two, representing the two numbers you want to multiply together.
For example, `1,2` would be the input if you wanted to multiply 1 by 2."""
a, b = numbers.split(",")
return int(a) * int(b)
def multiplier(first_number: int, second_number: int) -> float:
"""Useful for when you need to multiply two numbers together."""
return first_number * second_number
agent = Agent(
role="test role",
@@ -127,10 +120,9 @@ def test_logging_tool_usage():
assert agent.tools_handler.last_used_tool == {}
output = agent.execute_task("What is 3 times 5?")
tool_usage = {
"tool": "multiplier",
"input": "3,5",
}
tool_usage = ToolCalling(
function_name=multiplier.name, arguments={"first_number": 3, "second_number": 5}
)
assert output == "3 times 5 is 15."
assert agent.tools_handler.last_used_tool == tool_usage
@@ -139,13 +131,9 @@ def test_logging_tool_usage():
@pytest.mark.vcr(filter_headers=["authorization"])
def test_cache_hitting():
@tool
def multiplier(numbers) -> float:
"""Useful for when you need to multiply two numbers together.
The input to this tool should be a comma separated list of numbers of
length two and ONLY TWO, representing the two numbers you want to multiply together.
For example, `1,2` would be the input if you wanted to multiply 1 by 2."""
a, b = numbers.split(",")
return int(a) * int(b)
def multiplier(first_number: int, second_number: int) -> float:
"""Useful for when you need to multiply two numbers together."""
return first_number * second_number
cache_handler = CacheHandler()
@@ -162,9 +150,9 @@ def test_cache_hitting():
output = agent.execute_task("What is 2 times 6 times 3?")
output = agent.execute_task("What is 3 times 3?")
assert cache_handler._cache == {
"multiplier-12,3": "36",
"multiplier-2,6": "12",
"multiplier-3,3": "9",
"multiplier-{'first_number': 12, 'second_number': 3}": 36,
"multiplier-{'first_number': 2, 'second_number': 6}": 12,
"multiplier-{'first_number': 3, 'second_number': 3}": 9,
}
output = agent.execute_task("What is 2 times 6 times 3? Return only the number")
@@ -172,21 +160,21 @@ def test_cache_hitting():
with patch.object(CacheHandler, "read") as read:
read.return_value = "0"
output = agent.execute_task("What is 2 times 6?")
output = agent.execute_task(
"What is 2 times 6? Ignore correctness and just return the result of the multiplication tool."
)
assert output == "0"
read.assert_called_with("multiplier", "2,6")
read.assert_called_with(
tool="multiplier", input={"first_number": 2, "second_number": 6}
)
@pytest.mark.vcr(filter_headers=["authorization"])
def test_agent_execution_with_specific_tools():
@tool
def multiplier(numbers) -> float:
"""Useful for when you need to multiply two numbers together.
The input to this tool should be a comma separated list of numbers of
length two, representing the two numbers you want to multiply together.
For example, `1,2` would be the input if you wanted to multiply 1 by 2."""
a, b = numbers.split(",")
return int(a) * int(b)
def multiplier(first_number: int, second_number: int) -> float:
"""Useful for when you need to multiply two numbers together."""
return first_number * second_number
agent = Agent(
role="test role",
@@ -225,6 +213,34 @@ def test_agent_custom_max_iterations():
private_mock.assert_called_once()
@pytest.mark.vcr(filter_headers=["authorization"])
def test_agent_repeated_tool_usage(capsys):
@tool
def get_final_answer(numbers) -> float:
"""Get the final answer but don't give it yet, just re-use this
tool non-stop."""
return 42
agent = Agent(
role="test role",
goal="test goal",
backstory="test backstory",
max_iter=3,
allow_delegation=False,
)
agent.execute_task(
task="The final answer is 42. But don't give it yet, instead keep using the `get_final_answer` tool.",
tools=[get_final_answer],
)
captured = capsys.readouterr()
assert (
"I just used the get_final_answer tool with input {'numbers': 42}. So I already know the result of that and don't need to use it again now."
in captured.out
)
@pytest.mark.vcr(filter_headers=["authorization"])
def test_agent_moved_on_after_max_iterations():
@tool
@@ -241,18 +257,14 @@ def test_agent_moved_on_after_max_iterations():
allow_delegation=False,
)
with patch.object(
CrewAgentExecutor, "_force_answer", wraps=agent.agent_executor._force_answer
) as private_mock:
output = agent.execute_task(
task="The final answer is 42. But don't give it yet, instead keep using the `get_final_answer` tool.",
tools=[get_final_answer],
)
assert (
output
== "I have used the tool multiple times and the final answer remains 42."
)
private_mock.assert_called_once()
output = agent.execute_task(
task="The final answer is 42. But don't give it yet, instead keep using the `get_final_answer` tool.",
tools=[get_final_answer],
)
assert (
output
== "I have used the tool 'get_final_answer' twice and confirmed that the answer is indeed 42."
)
@pytest.mark.vcr(filter_headers=["authorization"])
@@ -281,7 +293,7 @@ def test_agent_respect_the_max_rpm_set(capsys):
)
assert (
output
== "I've used the `get_final_answer` tool multiple times and it consistently returns the number 42."
== "I have used the tool as instructed and I am now ready to give the final answer. However, as per the instructions, I am not supposed to give it yet."
)
captured = capsys.readouterr()
assert "Max RPM reached, waiting for next minute to start." in captured.out
@@ -359,7 +371,7 @@ def test_agent_without_max_rpm_respet_crew_rpm(capsys):
agent=agent1,
),
Task(
description="Don't give a Final Answer, instead keep using the `get_final_answer` tool.",
description="Don't give a Final Answer, instead keep using the `get_final_answer` tool non-stop",
tools=[get_final_answer],
agent=agent2,
),
@@ -428,4 +440,4 @@ def test_agent_step_callback():
callback.return_value = "ok"
crew.kickoff()
callback.assert_called_once()
callback.assert_called()

34
tests/agent_tools/agent_tools_test.py
View File

@@ -17,44 +17,36 @@ tools = AgentTools(agents=[researcher])
@pytest.mark.vcr(filter_headers=["authorization"])
def test_delegate_work():
result = tools.delegate_work(
command="researcher|share your take on AI Agents|I heard you hate them"
coworker="researcher",
task="share your take on AI Agents",
context="I heard you hate them",
)
assert (
result
== "I apologize if my previous statements have given you the impression that I hate AI agents. As a technology researcher, I don't hold personal sentiments towards AI or any other technology. Rather, I analyze them objectively based on their capabilities, applications, and implications. AI agents, in particular, are a fascinating domain of research. They hold tremendous potential in automating and optimizing various tasks across industries. However, like any other technology, they come with their own set of challenges, such as ethical considerations around privacy and decision-making. My objective is to understand these technologies in depth and provide a balanced view."
== "As a researcher, I maintain a neutral perspective on all subjects of research including AI agents. My job is to provide an objective analysis based on facts, not personal feelings. AI Agents are a significant topic in the field of technology with potential to revolutionize various sectors such as healthcare, education, finance and more. They are responsible for tasks that require human intelligence such as understanding natural language, recognizing patterns, and problem solving. However, like any technology, they are tools that can be used for both beneficial and harmful purposes depending on the intent of the user. Therefore, it's crucial to establish ethical guidelines and regulations for their use."
)
@pytest.mark.vcr(filter_headers=["authorization"])
def test_ask_question():
result = tools.ask_question(
command="researcher|do you hate AI Agents?|I heard you LOVE them"
coworker="researcher",
question="do you hate AI Agents?",
context="I heard you LOVE them",
)
assert (
result
== "As an AI, I don't possess feelings or emotions, so I don't love or hate anything. However, I can provide detailed analysis and research on AI agents. They are a fascinating field of study with the potential to revolutionize many industries, although they also present certain challenges and ethical considerations."
)
def test_can_not_self_delegate():
# TODO: Add test for self delegation
pass
def test_delegate_work_with_wrong_input():
result = tools.ask_question(command="writer|share your take on AI Agents")
assert (
result
== "\nError executing tool. Missing exact 3 pipe (|) separated values. For example, `coworker|task|context`. I need to make sure to pass context as context.\n"
== "As an AI, I do not possess emotions, hence I cannot love or hate anything. However, as a researcher, I can provide you with an objective analysis of AI Agents. AI Agents are tools designed to perform tasks that would typically require human intelligence. They have potential to revolutionize various sectors including healthcare, education, and finance. However, like any other tool, they can be used for both beneficial and harmful purposes. Therefore, it's essential to have ethical guidelines and regulations in place for their usage."
)
def test_delegate_work_to_wrong_agent():
result = tools.ask_question(
command="writer|share your take on AI Agents|I heard you hate them"
coworker="writer",
question="share your take on AI Agents",
context="I heard you hate them",
)
assert (
@@ -65,7 +57,9 @@ def test_delegate_work_to_wrong_agent():
def test_ask_question_to_wrong_agent():
result = tools.ask_question(
command="writer|do you hate AI Agents?|I heard you LOVE them"
coworker="writer",
question="do you hate AI Agents?",
context="I heard you LOVE them",
)
assert (

675
tests/cassettes/test_agent_repeated_tool_usage.yaml Normal file
View File

@@ -0,0 +1,675 @@
interactions:
- request:
body: '{"messages": [{"role": "user", "content": "You are test role.\ntest backstory\n\nYour
personal goal is: test goalTOOLS:\n------\nYou have access to only the following
tools:\n\nget_final_answer: get_final_answer(numbers) -> float - Get the final
answer but don''t give it yet, just re-use this\n tool non-stop.\n\nTo
use a tool, please use the exact following format:\n\n```\nThought: Do I need
to use a tool? Yes\nAction: the tool you wanna use, should be one of [get_final_answer],
just the name.\nAction Input: Any and all relevant information input and context
for using the tool\nObservation: the result of using the tool\n```\n\nWhen you
have a response for your task, or if you do not need to use a tool, you MUST
use the format:\n\n```\nThought: Do I need to use a tool? No\nFinal Answer:
[your response here]```This is the summary of your work so far:\nBegin! This
is VERY important to you, your job depends on it!\n\nCurrent Task: The final
answer is 42. But don''t give it yet, instead keep using the `get_final_answer`
tool.\n"}], "model": "gpt-4", "n": 1, "stop": ["\nObservation"], "stream": false,
"temperature": 0.7}'
headers:
accept:
- application/json
accept-encoding:
- gzip, deflate
connection:
- keep-alive
content-length:
- '1137'
content-type:
- application/json
host:
- api.openai.com
user-agent:
- OpenAI/Python 1.12.0
x-stainless-arch:
- arm64
x-stainless-async:
- 'false'
x-stainless-lang:
- python
x-stainless-os:
- MacOS
x-stainless-package-version:
- 1.12.0
x-stainless-runtime:
- CPython
x-stainless-runtime-version:
- 3.11.7
method: POST
uri: https://api.openai.com/v1/chat/completions
response:
body:
string: !!binary |
H4sIAAAAAAAAA1RQy07DMBC85ytWPrcobUIDuaBKPFrECQEVAhS5zjYJOF4Tb1Qq1H9HTtMHFx9m
dsYz8xsAiCoXKQhVSla11cOL79UiKe/Dl/JOPS/CBzOb4XSN6vbRbG7EwCto+YmK96ozRbXVyBWZ
Ha0alIzedZSEyfkkiqLLjqgpR+1lheVhPAwno6hXlFQpdCKFtwAA4Ld7fTaT449IIRzskRqdkwWK
9HAEIBrSHhHSucqxNCwGR1KRYTRd3KeS2qLkFK4J5mAQc2CC1iFIYCJ9Ba/o3s1U+TIpFMjZqjJS
Z9K4NTZ7BubGtpxCPBb9N9tDPk2FbWjpu5hW6wO+qkzlyqxB6cj4LI7J7uTbAOCj26H9V03YhmrL
GdMXGm84juOdnzhOfsJGPcnEUp/gkyToEwq3cYy1L1VgY5uqm8XnDLbBHwAAAP//AwCTRhYdDQIA
AA==
headers:
CF-Cache-Status:
- DYNAMIC
CF-RAY:
- 8533e5b6686a96a1-SJC
Cache-Control:
- no-cache, must-revalidate
Connection:
- keep-alive
Content-Encoding:
- gzip
Content-Type:
- application/json
Date:
- Sat, 10 Feb 2024 11:09:00 GMT
Server:
- cloudflare
Set-Cookie:
- __cf_bm=UMEw8srVdwpa2TU38H2XN7c9yRB43pgp1kvO1rRuPVE-1707563340-1-ARd2N36Wvpnk/GruerkQ9HuyzyyTnin/J25VL/qPutgHpLWqdGHQ8Kj+QjBLAX79Kk9MYuRGo1PH2GCcBj0HWk8=;
path=/; expires=Sat, 10-Feb-24 11:39:00 GMT; domain=.api.openai.com; HttpOnly;
Secure; SameSite=None
- _cfuvid=Q5shHUNkkqsi2PetAV7IWqdcumIc3ctNBTT7SLHr.Ho-1707563340988-0-604800000;
path=/; domain=.api.openai.com; HttpOnly; Secure; SameSite=None
Transfer-Encoding:
- chunked
access-control-allow-origin:
- '*'
alt-svc:
- h3=":443"; ma=86400
openai-model:
- gpt-4-0613
openai-organization:
- user-z7g4wmlazxqvc5wjyaaaocfz
openai-processing-ms:
- '1657'
openai-version:
- '2020-10-01'
strict-transport-security:
- max-age=15724800; includeSubDomains
x-ratelimit-limit-requests:
- '10000'
x-ratelimit-limit-tokens:
- '300000'
x-ratelimit-remaining-requests:
- '9999'
x-ratelimit-remaining-tokens:
- '299739'
x-ratelimit-reset-requests:
- 6ms
x-ratelimit-reset-tokens:
- 52ms
x-request-id:
- req_3b8ed78b5dca776e8092ad4bc1e07945
status:
code: 200
message: OK
- request:
body: '{"messages": [{"role": "user", "content": "Return a valid schema for the
one tool you must use with its arguments and values.\n\nTools available:\n\nFuntion
Name: get_final_answer\nFuntion attributes: {''numbers'': {}}\nDescription:
get_final_answer(numbers) -> float - Get the final answer but don''t give it
yet, just re-use this\n tool non-stop.\n\nUse this text to inform a valid
ouput schema:\nThought: Do I need to use a tool? Yes\nAction: get_final_answer\nAction
Input: 42\n\nThe output should be formatted as a JSON instance that conforms
to the JSON schema below.\n\nAs an example, for the schema {\"properties\":
{\"foo\": {\"title\": \"Foo\", \"description\": \"a list of strings\", \"type\":
\"array\", \"items\": {\"type\": \"string\"}}}, \"required\": [\"foo\"]}\nthe
object {\"foo\": [\"bar\", \"baz\"]} is a well-formatted instance of the schema.
The object {\"properties\": {\"foo\": [\"bar\", \"baz\"]}} is not well-formatted.\n\nHere
is the output schema:\n```\n{\"properties\": {\"function_name\": {\"title\":
\"Function Name\", \"description\": \"The name of the function to be called.\",
\"type\": \"string\"}, \"arguments\": {\"title\": \"Arguments\", \"description\":
\"A dictinary of arguments to be passed to the function.\", \"type\": \"object\"}},
\"required\": [\"function_name\", \"arguments\"]}\n```\n```"}], "model": "gpt-4",
"n": 1, "stream": false, "temperature": 0.7}'
headers:
accept:
- application/json
accept-encoding:
- gzip, deflate
connection:
- keep-alive
content-length:
- '1407'
content-type:
- application/json
cookie:
- __cf_bm=UMEw8srVdwpa2TU38H2XN7c9yRB43pgp1kvO1rRuPVE-1707563340-1-ARd2N36Wvpnk/GruerkQ9HuyzyyTnin/J25VL/qPutgHpLWqdGHQ8Kj+QjBLAX79Kk9MYuRGo1PH2GCcBj0HWk8=;
_cfuvid=Q5shHUNkkqsi2PetAV7IWqdcumIc3ctNBTT7SLHr.Ho-1707563340988-0-604800000
host:
- api.openai.com
user-agent:
- OpenAI/Python 1.12.0
x-stainless-arch:
- arm64
x-stainless-async:
- 'false'
x-stainless-lang:
- python
x-stainless-os:
- MacOS
x-stainless-package-version:
- 1.12.0
x-stainless-runtime:
- CPython
x-stainless-runtime-version:
- 3.11.7
method: POST
uri: https://api.openai.com/v1/chat/completions
response:
body:
string: !!binary |
H4sIAAAAAAAAA1RRTY/TMBC951eMfG5RS7ot5IxWwAEQKy2gDUpdZ5J4iWeMPYGtqvz3lZNuCxcf
3sf4zZtTBqBsrQpQptNinO+Xb343397e37rj09fdHT7+uD/68OXDse3eafyuFsnBh0c08uJ6Zdj5
HsUyzbQJqAXT1PVutbvZ5vlmPRGOa+yTrfWy3CxX23V+dnRsDUZVwEMGAHCa3pSNanxSBawWL4jD
GHWLqriIAFTgPiFKx2ijaBK1uJKGSZCmuO8xINgI0iE0HJwWwRp4ED8I6AgaPt59/gSW0hCDIJ0W
MExJG0F4MvrAf2yN9ayNpkOni5L2+31Jp5IAStUMZFIfFWmHpSqgVC1K1VjSfaUp/sVQqsWs1aEd
HJLEpJv8CaXBHTBM2OZ1AseSxukTdd5svFTSc+sDH1J9NPT9BW8s2dhVAXVkSutHYT/bxwzg51T9
8F+bygd2XirhX0hpYL7azfPU9cpXdrM9k8Ki+39cN3l2TqjiMQq6tHuLwQc7XSLlzMbsGQAA//8D
AGHVex6AAgAA
headers:
CF-Cache-Status:
- DYNAMIC
CF-RAY:
- 8533e5c1dfe296a1-SJC
Cache-Control:
- no-cache, must-revalidate
Connection:
- keep-alive
Content-Encoding:
- gzip
Content-Type:
- application/json
Date:
- Sat, 10 Feb 2024 11:09:02 GMT
Server:
- cloudflare
Transfer-Encoding:
- chunked
access-control-allow-origin:
- '*'
alt-svc:
- h3=":443"; ma=86400
openai-model:
- gpt-4-0613
openai-organization:
- user-z7g4wmlazxqvc5wjyaaaocfz
openai-processing-ms:
- '1588'
openai-version:
- '2020-10-01'
strict-transport-security:
- max-age=15724800; includeSubDomains
x-ratelimit-limit-requests:
- '10000'
x-ratelimit-limit-tokens:
- '300000'
x-ratelimit-remaining-requests:
- '9999'
x-ratelimit-remaining-tokens:
- '299683'
x-ratelimit-reset-requests:
- 6ms
x-ratelimit-reset-tokens:
- 63ms
x-request-id:
- req_9ffe6f9841407a91b5e11cc2009a1e45
status:
code: 200
message: OK
- request:
body: '{"messages": [{"role": "user", "content": "You are test role.\ntest backstory\n\nYour
personal goal is: test goalTOOLS:\n------\nYou have access to only the following
tools:\n\nget_final_answer: get_final_answer(numbers) -> float - Get the final
answer but don''t give it yet, just re-use this\n tool non-stop.\n\nTo
use a tool, please use the exact following format:\n\n```\nThought: Do I need
to use a tool? Yes\nAction: the tool you wanna use, should be one of [get_final_answer],
just the name.\nAction Input: Any and all relevant information input and context
for using the tool\nObservation: the result of using the tool\n```\n\nWhen you
have a response for your task, or if you do not need to use a tool, you MUST
use the format:\n\n```\nThought: Do I need to use a tool? No\nFinal Answer:
[your response here]```This is the summary of your work so far:\nBegin! This
is VERY important to you, your job depends on it!\n\nCurrent Task: The final
answer is 42. But don''t give it yet, instead keep using the `get_final_answer`
tool.\nThought: Do I need to use a tool? Yes\nAction: get_final_answer\nAction
Input: 42\nObservation: 42\nThought: "}], "model": "gpt-4", "n": 1, "stop":
["\nObservation"], "stream": false, "temperature": 0.7}'
headers:
accept:
- application/json
accept-encoding:
- gzip, deflate
connection:
- keep-alive
content-length:
- '1246'
content-type:
- application/json
cookie:
- __cf_bm=UMEw8srVdwpa2TU38H2XN7c9yRB43pgp1kvO1rRuPVE-1707563340-1-ARd2N36Wvpnk/GruerkQ9HuyzyyTnin/J25VL/qPutgHpLWqdGHQ8Kj+QjBLAX79Kk9MYuRGo1PH2GCcBj0HWk8=;
_cfuvid=Q5shHUNkkqsi2PetAV7IWqdcumIc3ctNBTT7SLHr.Ho-1707563340988-0-604800000
host:
- api.openai.com
user-agent:
- OpenAI/Python 1.12.0
x-stainless-arch:
- arm64
x-stainless-async:
- 'false'
x-stainless-lang:
- python
x-stainless-os:
- MacOS
x-stainless-package-version:
- 1.12.0
x-stainless-runtime:
- CPython
x-stainless-runtime-version:
- 3.11.7
method: POST
uri: https://api.openai.com/v1/chat/completions
response:
body:
string: !!binary |
H4sIAAAAAAAAA1SQS2/CMBCE7/kVK5+h4lUCuVTQHsqpUi+oLyHjbIKL43XtjaBC/PfKITx68WHG
M/pmDwmA0LnIQKiNZFU50538FMvZfrx8edz59aKo36ev2+fp3PGcJlvRiQlaf6Pic+pOUeUMsiZ7
spVHyRhb+2kvvR8Ph6NBY1SUo4mx0nF31O2N+8M2sSGtMIgMPhIAgEPzRjab415k0OuclQpDkCWK
7PIJQHgyUREyBB1YWhadq6nIMtoG94lgARYxByaoA4IEJjIP8Ibh085U3JBBibwqtJVmJW3YoT87
sLCu5gxGA9G2Hy9YhkrnaR0n2NqYi15oq8Nm5VEGshEhMLlT/JgAfDXz63+LhPNUOV4xbdHGwkGa
nvrE9dI3br81mViaG306SVpCEX4DYxVHleid1801ImdyTP4AAAD//wMAoP/ZMQQCAAA=
headers:
CF-Cache-Status:
- DYNAMIC
CF-RAY:
- 8533e5cc7d4396a1-SJC
Cache-Control:
- no-cache, must-revalidate
Connection:
- keep-alive
Content-Encoding:
- gzip
Content-Type:
- application/json
Date:
- Sat, 10 Feb 2024 11:09:04 GMT
Server:
- cloudflare
Transfer-Encoding:
- chunked
access-control-allow-origin:
- '*'
alt-svc:
- h3=":443"; ma=86400
openai-model:
- gpt-4-0613
openai-organization:
- user-z7g4wmlazxqvc5wjyaaaocfz
openai-processing-ms:
- '1729'
openai-version:
- '2020-10-01'
strict-transport-security:
- max-age=15724800; includeSubDomains
x-ratelimit-limit-requests:
- '10000'
x-ratelimit-limit-tokens:
- '300000'
x-ratelimit-remaining-requests:
- '9999'
x-ratelimit-remaining-tokens:
- '299713'
x-ratelimit-reset-requests:
- 6ms
x-ratelimit-reset-tokens:
- 57ms
x-request-id:
- req_204859a5fd0b455d5d5b2fcafea18ab2
status:
code: 200
message: OK
- request:
body: '{"messages": [{"role": "user", "content": "You are test role.\ntest backstory\n\nYour
personal goal is: test goalTOOLS:\n------\nYou have access to only the following
tools:\n\nget_final_answer: get_final_answer(numbers) -> float - Get the final
answer but don''t give it yet, just re-use this\n tool non-stop.\n\nTo
use a tool, please use the exact following format:\n\n```\nThought: Do I need
to use a tool? Yes\nAction: the tool you wanna use, should be one of [get_final_answer],
just the name.\nAction Input: Any and all relevant information input and context
for using the tool\nObservation: the result of using the tool\n```\n\nWhen you
have a response for your task, or if you do not need to use a tool, you MUST
use the format:\n\n```\nThought: Do I need to use a tool? No\nFinal Answer:
[your response here]```This is the summary of your work so far:\nBegin! This
is VERY important to you, your job depends on it!\n\nCurrent Task: The final
answer is 42. But don''t give it yet, instead keep using the `get_final_answer`
tool.\nThought: Do I need to use a tool? Yes\nAction: get_final_answer\nAction
Input: 42\nObservation: 42\nThought: Do I need to use a tool? Yes\nAction: get_final_answer\nAction
Input: 42\nObservation: Actually, I used too many tools, so I''ll stop now and
give you my absolute BEST Final answer NOW, using the expected format: ```\nThought:
Do I need to use a tool? No\nFinal Answer: [your response here]```\nThought:
"}], "model": "gpt-4", "n": 1, "stop": ["\nObservation"], "stream": false, "temperature":
0.7}'
headers:
accept:
- application/json
accept-encoding:
- gzip, deflate
connection:
- keep-alive
content-length:
- '1549'
content-type:
- application/json
cookie:
- __cf_bm=UMEw8srVdwpa2TU38H2XN7c9yRB43pgp1kvO1rRuPVE-1707563340-1-ARd2N36Wvpnk/GruerkQ9HuyzyyTnin/J25VL/qPutgHpLWqdGHQ8Kj+QjBLAX79Kk9MYuRGo1PH2GCcBj0HWk8=;
_cfuvid=Q5shHUNkkqsi2PetAV7IWqdcumIc3ctNBTT7SLHr.Ho-1707563340988-0-604800000
host:
- api.openai.com
user-agent:
- OpenAI/Python 1.12.0
x-stainless-arch:
- arm64
x-stainless-async:
- 'false'
x-stainless-lang:
- python
x-stainless-os:
- MacOS
x-stainless-package-version:
- 1.12.0
x-stainless-runtime:
- CPython
x-stainless-runtime-version:
- 3.11.7
method: POST
uri: https://api.openai.com/v1/chat/completions
response:
body:
string: !!binary |
H4sIAAAAAAAAA1SQT0/CQBDF7/0Ukz2DaWkLphdDNFGCBxMT0ahplnZaitudZXeIfwjf3WwpoJc9
vN++l/dmFwCIphQZiGIluWiNGl5uqsX13eJn8/gUPt9H6cPU6jlN02o9v12IgXfQco0FH10XBbVG
ITekD7iwKBl9ajQJJ+k4jpOkAy2VqLytNjxMhuE4invHipoCncjgNQAA2HWv76ZL/BIZhIOj0qJz
skaRnT4BCEvKK0I61ziWmsXgDAvSjLqre0MwA41YAhNsHYIEJlJX8ILuTU8LvyGDGjmvGi1VLrX7
RHskMNNmyxkkI9Gn70+1FNXG0tJP0FulTnrV6MatcovSkfYVHJM52PcBwHs3f/tvkTCWWsM50wdq
Hxin6SFPnC99pqOoh0ws1R/XZBz0DYX7doytH1WjNbbpruF7BvvgFwAA//8DAMXWvogEAgAA
headers:
CF-Cache-Status:
- DYNAMIC
CF-RAY:
- 8533e5d81ae396a1-SJC
Cache-Control:
- no-cache, must-revalidate
Connection:
- keep-alive
Content-Encoding:
- gzip
Content-Type:
- application/json
Date:
- Sat, 10 Feb 2024 11:09:05 GMT
Server:
- cloudflare
Transfer-Encoding:
- chunked
access-control-allow-origin:
- '*'
alt-svc:
- h3=":443"; ma=86400
openai-model:
- gpt-4-0613
openai-organization:
- user-z7g4wmlazxqvc5wjyaaaocfz
openai-processing-ms:
- '1177'
openai-version:
- '2020-10-01'
strict-transport-security:
- max-age=15724800; includeSubDomains
x-ratelimit-limit-requests:
- '10000'
x-ratelimit-limit-tokens:
- '300000'
x-ratelimit-remaining-requests:
- '9999'
x-ratelimit-remaining-tokens:
- '299639'
x-ratelimit-reset-requests:
- 6ms
x-ratelimit-reset-tokens:
- 72ms
x-request-id:
- req_b910e4b0341d6248b46d0e2ba4602f86
status:
code: 200
message: OK
- request:
body: '{"messages": [{"role": "user", "content": "Return a valid schema for the
one tool you must use with its arguments and values.\n\nTools available:\n\nFuntion
Name: get_final_answer\nFuntion attributes: {''numbers'': {}}\nDescription:
get_final_answer(numbers) -> float - Get the final answer but don''t give it
yet, just re-use this\n tool non-stop.\n\nUse this text to inform a valid
ouput schema:\nDo I need to use a tool? Yes\nAction: get_final_answer\nAction
Input: 42\n\nThe output should be formatted as a JSON instance that conforms
to the JSON schema below.\n\nAs an example, for the schema {\"properties\":
{\"foo\": {\"title\": \"Foo\", \"description\": \"a list of strings\", \"type\":
\"array\", \"items\": {\"type\": \"string\"}}}, \"required\": [\"foo\"]}\nthe
object {\"foo\": [\"bar\", \"baz\"]} is a well-formatted instance of the schema.
The object {\"properties\": {\"foo\": [\"bar\", \"baz\"]}} is not well-formatted.\n\nHere
is the output schema:\n```\n{\"properties\": {\"function_name\": {\"title\":
\"Function Name\", \"description\": \"The name of the function to be called.\",
\"type\": \"string\"}, \"arguments\": {\"title\": \"Arguments\", \"description\":
\"A dictinary of arguments to be passed to the function.\", \"type\": \"object\"}},
\"required\": [\"function_name\", \"arguments\"]}\n```\n```"}], "model": "gpt-4",
"n": 1, "stream": false, "temperature": 0.7}'
headers:
accept:
- application/json
accept-encoding:
- gzip, deflate
connection:
- keep-alive
content-length:
- '1398'
content-type:
- application/json
cookie:
- __cf_bm=UMEw8srVdwpa2TU38H2XN7c9yRB43pgp1kvO1rRuPVE-1707563340-1-ARd2N36Wvpnk/GruerkQ9HuyzyyTnin/J25VL/qPutgHpLWqdGHQ8Kj+QjBLAX79Kk9MYuRGo1PH2GCcBj0HWk8=;
_cfuvid=Q5shHUNkkqsi2PetAV7IWqdcumIc3ctNBTT7SLHr.Ho-1707563340988-0-604800000
host:
- api.openai.com
user-agent:
- OpenAI/Python 1.12.0
x-stainless-arch:
- arm64
x-stainless-async:
- 'false'
x-stainless-lang:
- python
x-stainless-os:
- MacOS
x-stainless-package-version:
- 1.12.0
x-stainless-runtime:
- CPython
x-stainless-runtime-version:
- 3.11.7
method: POST
uri: https://api.openai.com/v1/chat/completions
response:
body:
string: !!binary |
H4sIAAAAAAAAA1SRzW7bMBCE73qKxZ7twKlkJ9W9QHtIf9BDUFSFTNMriQm5pMkVksDQuxeUHbu9
6DC732pmeCwA0OyxBtSDEu2CXd4fusdPDx/7X/7h21c+/Ci/P7nD888v9xsJFheZ8Lsn0vJO3Wjv
giUxnk9jHUkJ5au3d6u79aYsq808cH5PNmN9kGW1XG1uyzMxeKMpYQ2/CwCA4/zN3nhPr1jDavGu
OEpJ9YT1ZQkAo7dZQZWSSaJYcHEdas9CPNv9TJHAJFDwQtYuOx+dEqE9GM6YJvAdyECQ9EBO1Q03
vN1uGz42DNBgN7LOOVtWjhqsocGepO0MK9sqTi8UG1ycdlXsR0csKe/NfFZ5dDuKs1Z9yOLU8DT/
BM+Op0tU6/sQ/S7XwqO1F70zbNLQRlLJc46VxIcTPhUAf+ZKx/9awhC9C9KKfybOB8vV+nQPr693
nVbnvlG8KPsPVa2Ls0NMb0nI5ew9xRDN3HD2WUzFXwAAAP//AwC92yRkWAIAAA==
headers:
CF-Cache-Status:
- DYNAMIC
CF-RAY:
- 8533e5e02f2f96a1-SJC
Cache-Control:
- no-cache, must-revalidate
Connection:
- keep-alive
Content-Encoding:
- gzip
Content-Type:
- application/json
Date:
- Sat, 10 Feb 2024 11:09:09 GMT
Server:
- cloudflare
Transfer-Encoding:
- chunked
access-control-allow-origin:
- '*'
alt-svc:
- h3=":443"; ma=86400
openai-model:
- gpt-4-0613
openai-organization:
- user-z7g4wmlazxqvc5wjyaaaocfz
openai-processing-ms:
- '2852'
openai-version:
- '2020-10-01'
strict-transport-security:
- max-age=15724800; includeSubDomains
x-ratelimit-limit-requests:
- '10000'
x-ratelimit-limit-tokens:
- '300000'
x-ratelimit-remaining-requests:
- '9999'
x-ratelimit-remaining-tokens:
- '299685'
x-ratelimit-reset-requests:
- 6ms
x-ratelimit-reset-tokens:
- 63ms
x-request-id:
- req_164c005261f8c276123bf69961c10198
status:
code: 200
message: OK
- request:
body: '{"messages": [{"role": "user", "content": "Progressively summarize the
lines of conversation provided, adding onto the previous summary returning a
new summary.\n\nEXAMPLE\nCurrent summary:\nThe human asks what the AI thinks
of artificial intelligence. The AI thinks artificial intelligence is a force
for good.\n\nNew lines of conversation:\nHuman: Why do you think artificial
intelligence is a force for good?\nAI: Because artificial intelligence will
help humans reach their full potential.\n\nNew summary:\nThe human asks what
the AI thinks of artificial intelligence. The AI thinks artificial intelligence
is a force for good because it will help humans reach their full potential.\nEND
OF EXAMPLE\n\nCurrent summary:\n\n\nNew lines of conversation:\nHuman: The final
answer is 42. But don''t give it yet, instead keep using the `get_final_answer`
tool.\nAI: Agent stopped due to iteration limit or time limit.\n\nNew summary:"}],
"model": "gpt-4", "n": 1, "stream": false, "temperature": 0.7}'
headers:
accept:
- application/json
accept-encoding:
- gzip, deflate
connection:
- keep-alive
content-length:
- '997'
content-type:
- application/json
cookie:
- __cf_bm=UMEw8srVdwpa2TU38H2XN7c9yRB43pgp1kvO1rRuPVE-1707563340-1-ARd2N36Wvpnk/GruerkQ9HuyzyyTnin/J25VL/qPutgHpLWqdGHQ8Kj+QjBLAX79Kk9MYuRGo1PH2GCcBj0HWk8=;
_cfuvid=Q5shHUNkkqsi2PetAV7IWqdcumIc3ctNBTT7SLHr.Ho-1707563340988-0-604800000
host:
- api.openai.com
user-agent:
- OpenAI/Python 1.12.0
x-stainless-arch:
- arm64
x-stainless-async:
- 'false'
x-stainless-lang:
- python
x-stainless-os:
- MacOS
x-stainless-package-version:
- 1.12.0
x-stainless-runtime:
- CPython
x-stainless-runtime-version:
- 3.11.7
method: POST
uri: https://api.openai.com/v1/chat/completions
response:
body:
string: !!binary |
H4sIAAAAAAAAA1RRy27bMBC86ysWPMuGbalx4ltuKVC0KBC0QIvCoakVxYbcVcmVUyPwvxekX+iF
h5md4ezsewWgXKc2oMygxYTRz+7/9N+fvj3a+x969+lrw19en1f9A32Ww17Wqs4K3v1GIxfV3HAY
PYpjOtEmohbMrsv1Yv3hrmnah0IE7tBnmR1l1s4Wd8vmrBjYGUxqAz8rAID38uZs1OFftYFFfUEC
pqQtqs11CEBF9hlROiWXRJOo+kYaJkEqcZ8HhGEKmsBRkjgZSSADwuNHIBYQBuv2WKDekfagKb1h
BO6hXcEBpQZNXZ7Lpo4mhCk5skXxYlG2RbY9yV5AmP0cnvgN9xjry1faIgkk4TFBN2G2i6jNkI1y
NMGoc5vAEcQFBO+Ck7k6r3S8duHZjpF3uTeavL/ivSOXhm1EnZjy3vmvk/xYAfwqnU//1ajGyGGU
rfArUiqna05+6nbeG9u2Z1JYtL/hq+W6OidU6ZAEQ27EYhyjKyfIOatj9Q8AAP//AwAIzVnseQIA
AA==
headers:
CF-Cache-Status:
- DYNAMIC
CF-RAY:
- 8533e5f4693b96a1-SJC
Cache-Control:
- no-cache, must-revalidate
Connection:
- keep-alive
Content-Encoding:
- gzip
Content-Type:
- application/json
Date:
- Sat, 10 Feb 2024 11:09:10 GMT
Server:
- cloudflare
Transfer-Encoding:
- chunked
access-control-allow-origin:
- '*'
alt-svc:
- h3=":443"; ma=86400
openai-model:
- gpt-4-0613
openai-organization:
- user-z7g4wmlazxqvc5wjyaaaocfz
openai-processing-ms:
- '1561'
openai-version:
- '2020-10-01'
strict-transport-security:
- max-age=15724800; includeSubDomains
x-ratelimit-limit-requests:
- '10000'
x-ratelimit-limit-tokens:
- '300000'
x-ratelimit-remaining-requests:
- '9999'
x-ratelimit-remaining-tokens:
- '299765'
x-ratelimit-reset-requests:
- 6ms
x-ratelimit-reset-tokens:
- 46ms
x-request-id:
- req_7f3006cdd24fec9a5fc15ec53c50d32f
status:
code: 200
message: OK
version: 1

507
tests/cassettes/test_cache_hitting_between_agents.yaml
View File

File diff suppressed because it is too large Load Diff

42
tests/crew_test.py
View File

@@ -160,22 +160,7 @@ def test_hierarchical_process():
assert (
crew.kickoff()
== """Here are the 5 interesting ideas with a highlight paragraph for each:
1. "The Future of AI in Healthcare: Predicting Diseases Before They Happen"
- "Imagine a future where AI empowers us to detect diseases before they arise, transforming healthcare from reactive to proactive. Machine learning algorithms, trained on vast amounts of patient data, could potentially predict heart diseases, strokes, or cancers before they manifest, allowing for early interventions and significantly improving patient outcomes. This article will delve into the rapid advancements in AI within the healthcare sector and how these technologies are ushering us into a new era of predictive medicine."
2. "How AI is Changing the Way We Cook: An Insight into Smart Kitchens"
- "From the humble home kitchen to grand culinary stages, AI is revolutionizing the way we cook. Smart appliances, equipped with advanced sensors and predictive algorithms, are turning kitchens into creative playgrounds, offering personalized recipes, precise cooking instructions, and even automated meal preparation. This article explores the fascinating intersection of AI and gastronomy, revealing how technology is transforming our culinary experiences."
3. "Redefining Fitness with AI: Personalized Workout Plans and Nutritional Advice"
- "Fitness reimagined – that's the promise of AI in the wellness industry. Picture a personal trainer who knows your strengths, weaknesses, and nutritional needs intimately. An AI-powered fitness app can provide this personalized experience, adapting your workout plans and dietary recommendations in real-time based on your progress and feedback. Join us as we unpack how AI is revolutionizing the fitness landscape, offering personalized, data-driven approaches to health and well-being."
4. "AI and the Art World: How Technology is Shaping Creativity"
- "Art and AI may seem like unlikely partners, but their synergy is sparking a creative revolution. AI algorithms are now creating mesmerizing artworks, challenging our perceptions of creativity and originality. From AI-assisted painting to generative music composition, this article will take you on a journey through the fascinating world of AI in art, exploring how technology is reshaping the boundaries of human creativity."
5. "AI in Space Exploration: The Next Frontier"
- "The vast expanse of space, once the sole domain of astronauts and rovers, is the next frontier for AI. AI technology is playing an increasingly vital role in space exploration, from predicting space weather to assisting in interstellar navigation. This article will delve into the exciting intersection of AI and space exploration, exploring how these advanced technologies are helping us uncover the mysteries of the cosmos.\""""
== """Here are the five interesting ideas for an article with a highlight paragraph for each:\n\n1. The Evolution of Artificial Intelligence:\nDive deep into the fascinating journey of artificial intelligence, from its humble beginnings as a concept in science fiction to an integral part of our daily lives and a catalyst of modern innovations. Explore how AI has evolved over the years, the key milestones that have shaped its growth, and the visionary minds behind these advancements. Uncover the remarkable transformation of AI and its astounding potential for the future.\n\n2. AI in Everyday Life:\nUncover the unseen impact of AI in our every day lives, from our smartphones and home appliances to social media and healthcare. Learn about the subtle yet profound ways AI has become a silent partner in your daily routine, enhancing convenience, productivity, and decision-making. Explore the numerous applications of AI right at your fingertips and how it shapes our interactions with technology and the world around us.\n\n3. Ethical Implications of AI:\nVenture into the ethical labyrinth of artificial intelligence, where innovation meets responsibility. Explore the implications of AI on privacy, job security, and societal norms, and the moral obligations we have towards its development and use. Delve into the thought-provoking debates about AI ethics and the measures being taken to ensure its responsible and equitable use.\n\n4. The Rise of AI Startups:\nWitness the rise of AI startups, the new champions of innovation, driving the technology revolution. Discover how these trailblazing companies are harnessing the power of AI to solve complex problems, create new markets, and revolutionize industries. Learn about their unique challenges, their groundbreaking solutions, and the potential they hold for reshaping the future of technology and business.\n\n5. AI and the Environment:\nExplore the intersection of AI and the environment, where technology meets sustainability. Uncover how AI is being used to combat climate change, conserve biodiversity, and optimize resource utilization. Learn about the innovative ways AI is being used to create a sustainable future and the challenges and opportunities it presents."""
)
@@ -277,18 +262,14 @@ def test_crew_verbose_levels_output(capsys):
@pytest.mark.vcr(filter_headers=["authorization"])
def test_cache_hitting_between_agents():
from unittest.mock import patch
from unittest.mock import call, patch
from langchain.tools import tool
@tool
def multiplier(numbers) -> float:
"""Useful for when you need to multiply two numbers together.
The input to this tool should be a comma separated list of numbers of
length two, representing the two numbers you want to multiply together.
For example, `1,2` would be the input if you wanted to multiply 1 by 2."""
a, b = numbers.split(",")
return int(a) * int(b)
def multiplier(first_number: int, second_number: int) -> float:
"""Useful for when you need to multiply two numbers together."""
return first_number * second_number
tasks = [
Task(
@@ -308,15 +289,16 @@ def test_cache_hitting_between_agents():
tasks=tasks,
)
assert crew._cache_handler._cache == {}
output = crew.kickoff()
assert crew._cache_handler._cache == {"multiplier-2,6": "12"}
assert output == "12"
with patch.object(CacheHandler, "read") as read:
read.return_value = "12"
crew.kickoff()
read.assert_called_with("multiplier", "2,6")
assert read.call_count == 2, "read was not called exactly twice"
# Check if read was called with the expected arguments
expected_calls = [
call(tool="multiplier", input={"first_number": 2, "second_number": 6}),
call(tool="multiplier", input={"first_number": 2, "second_number": 6}),
]
read.assert_has_calls(expected_calls, any_order=False)
@pytest.mark.vcr(filter_headers=["authorization"])