andre/crewAI
1
0
Fork 0
mirror of https://github.com/crewAIInc/crewAI.git synced 2026-01-09 16:18:30 +00:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity

adding function calling llm support

Browse Source
This commit is contained in:
João Moura
2024-02-13 02:57:12 -08:00
parent 2410d0c531
commit 55c0c186d1
11 changed files with 4378 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
docs/core-concepts/Agents.md
View File

@@ -20,7 +20,9 @@ description: What are crewAI Agents and how to use them.
| **Role** | Defines the agent's function within the crew. It determines the kind of tasks the agent is best suited for. |
| **Goal** | The individual objective that the agent aims to achieve. It guides the agent's decision-making process. |
| **Backstory** | Provides context to the agent's role and goal, enriching the interaction and collaboration dynamics. |
| **LLM** | The language model used by the agent to process and generate text. |
| **Tools** | Set of capabilities or functions that the agent can use to perform tasks. Tools can be shared or exclusive to specific agents. |
| **Function Calling LLM** | The language model used by this agent to call functions, if none is passed the same main llm for each agent will be used. |
| **Max Iter** | The maximum number of iterations the agent can perform before forced to give its best answer |
| **Max RPM** | The maximum number of requests per minute the agent can perform to avoid rate limits |
| **Verbose** | This allow you to actually see what is going on during the Crew execution. |
@@ -47,6 +49,8 @@ agent = Agent(
You're currently working on a project to analyze the
performance of our marketing campaigns.""",
tools=[my_tool1, my_tool2],
llm=my_llm,
function_calling_llm=my_llm,
max_iter=10,
max_rpm=10,
verbose=True,

1
docs/core-concepts/Crews.md
View File

@@ -16,6 +16,7 @@ description: Understanding and utilizing crews in the crewAI framework.
| **Process** | The process flow (e.g., sequential, hierarchical) the crew follows. |
| **Verbose** | The verbosity level for logging during execution. |
| **Manager LLM** | The language model used by the manager agent in a hierarchical process. |
| **Function Calling LLM** | The language model used by all agensts in the crew to call functions, if none is passed the same main llm for each agent will be used. |
| **Config** | Configuration settings for the crew. |
| **Max RPM** | Maximum requests per minute the crew adheres to during execution. |
| **Language** | Language setting for the crew's operation. |

8
src/crewai/agent.py
View File

@@ -36,6 +36,7 @@ class Agent(BaseModel):
goal: The objective of the agent.
backstory: The backstory of the agent.
llm: The language model that will run the agent.
function_calling_llm: The language model that will the tool calling for this agent, it overrides the crew function_calling_llm.
max_iter: Maximum number of iterations for an agent to execute a task.
memory: Whether the agent should have memory or not.
max_rpm: Maximum number of requests per minute for the agent execution to be respected.
@@ -98,6 +99,9 @@ class Agent(BaseModel):
),
description="Language model that will run the agent.",
)
function_calling_llm: Optional[Any] = Field(
description="Language model that will run the agent.", default=None
)
@field_validator("id", mode="before")
@classmethod
@@ -140,7 +144,6 @@ class Agent(BaseModel):
Returns:
Output of the agent
"""
task_prompt = task.prompt()
if context:
@@ -151,7 +154,7 @@ class Agent(BaseModel):
tools = tools or self.tools
self.agent_executor.tools = tools
self.agent_executor.task = task
self.agent_executor.tools_description = (render_text_description(tools),)
self.agent_executor.tools_description = render_text_description(tools)
self.agent_executor.tools_names = self.__tools_names(tools)
result = self.agent_executor.invoke(
@@ -208,6 +211,7 @@ class Agent(BaseModel):
"max_iterations": self.max_iter,
"step_callback": self.step_callback,
"tools_handler": self.tools_handler,
"function_calling_llm": self.function_calling_llm,
}
if self._rpm_controller:

2
src/crewai/agents/executor.py
View File

@@ -24,6 +24,7 @@ class CrewAgentExecutor(AgentExecutor):
task: Any = None
tools_description: str = ""
tools_names: str = ""
function_calling_llm: Any = None
request_within_rpm_limit: Any = None
tools_handler: InstanceOf[ToolsHandler] = None
max_iterations: Optional[int] = 15
@@ -194,6 +195,7 @@ class CrewAgentExecutor(AgentExecutor):
tools=self.tools,
tools_description=self.tools_description,
tools_names=self.tools_names,
function_calling_llm=self.function_calling_llm,
llm=self.llm,
task=self.task,
).use(agent_action.log)

4
src/crewai/agents/tools_handler.py
View File

@@ -22,9 +22,9 @@ class ToolsHandler:
def on_tool_end(self, calling: ToolCalling, output: str) -> Any:
"""Run when tool ends running."""
if self.last_used_tool.function_name != CacheTools().name:
if self.last_used_tool.tool_name != CacheTools().name:
self.cache.add(
tool=calling.function_name,
tool=calling.tool_name,
input=calling.arguments,
output=output,
)

10
src/crewai/crew.py
View File

@@ -32,6 +32,7 @@ class Crew(BaseModel):
tasks: List of tasks assigned to the crew.
agents: List of agents part of this crew.
manager_llm: The language model that will run manager agent.
function_calling_llm: The language model that will run the tool calling for all the agents.
process: The process flow that the crew will follow (e.g., sequential).
verbose: Indicates the verbosity level for logging during execution.
config: Configuration settings for the crew.
@@ -60,6 +61,9 @@ class Crew(BaseModel):
manager_llm: Optional[Any] = Field(
description="Language model that will run the agent.", default=None
)
function_calling_llm: Optional[Any] = Field(
description="Language model that will run the agent.", default=None
)
config: Optional[Union[Json, Dict[str, Any]]] = Field(default=None)
id: UUID4 = Field(default_factory=uuid.uuid4, frozen=True)
share_crew: Optional[bool] = Field(default=False)
@@ -176,7 +180,11 @@ class Crew(BaseModel):
for agent in self.agents:
agent.i18n = I18N(language=self.language)
if (self.step_callback) and (not agent.step_callback):
if not agent.function_calling_llm:
agent.function_calling_llm = self.function_calling_llm
agent.create_agent_executor()
if not agent.step_callback:
agent.step_callback = self.step_callback
agent.create_agent_executor()

27
src/crewai/tools/tool_usage.py
View File

@@ -41,6 +41,7 @@ class ToolUsage:
tools_names: str,
task: Any,
llm: Any,
function_calling_llm: Any,
) -> None:
self._i18n: I18N = I18N()
self._printer: Printer = Printer()
@@ -54,6 +55,7 @@ class ToolUsage:
self.tools = tools
self.task = task
self.llm = llm
self.function_calling_llm = function_calling_llm
def use(self, tool_string: str):
calling = self._tool_calling(tool_string)
@@ -79,7 +81,9 @@ class ToolUsage:
try:
result = self._i18n.errors("task_repeated_usage").format(
tool=calling.tool_name,
tool_input=", ".join(calling.arguments.values()),
tool_input=", ".join(
[str(arg) for arg in calling.arguments.values()]
),
)
self._printer.print(content=f"\n\n{result}\n", color="yellow")
self._telemetry.tool_repeated_usage(
@@ -138,7 +142,9 @@ class ToolUsage:
self, calling: Union[ToolCalling, InstructorToolCalling]
) -> None:
if last_tool_usage := self.tools_handler.last_used_tool:
return calling == last_tool_usage
return (calling.tool_name == last_tool_usage.tool_name) and (
calling.arguments == last_tool_usage.arguments
)
def _select_tool(self, tool_name: str) -> BaseTool:
for tool in self.tools:
@@ -175,15 +181,18 @@ class ToolUsage:
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
):
llm = self.function_calling_llm or self.llm
if (isinstance(llm, ChatOpenAI)) and (llm.openai_api_base == None):
print("CARALHOooooooooooo")
print(llm)
print("CARALHOooooooooooo")
client = instructor.patch(
self.llm.client._client,
llm.client._client,
mode=instructor.Mode.FUNCTIONS,
)
calling = client.chat.completions.create(
model=self.llm.model_name,
model=llm.model_name,
messages=[
{
"role": "system",
@@ -220,13 +229,13 @@ class ToolUsage:
""",
},
)
chain = prompt | self.llm | parser
chain = prompt | 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)
self._telemetry.tool_usage_error(llm=llm)
return ToolUsageErrorException(self._i18n.errors("tool_usage_error"))
return self._tool_calling(tool_string)

32
tests/agent_test.py
View File

@@ -543,3 +543,35 @@ def test_agent_step_callback():
callback.return_value = "ok"
crew.kickoff()
callback.assert_called()
@pytest.mark.vcr(filter_headers=["authorization"])
def test_agent_function_calling_llm():
from langchain_openai import ChatOpenAI
llm = ChatOpenAI(model="gpt-3.5")
with patch.object(llm.client, "create", wraps=llm.client.create) as private_mock:
@tool
def learn_about_AI(topic) -> float:
"""Useful for when you need to learn about AI to write an paragraph about it."""
return "AI is a very broad field."
agent1 = Agent(
role="test role",
goal="test goal",
backstory="test backstory",
tools=[learn_about_AI],
function_calling_llm=llm,
)
essay = Task(
description="Write and then review an small paragraph on AI until it's AMAZING",
agent=agent1,
)
tasks = [essay]
crew = Crew(agents=[agent1], tasks=tasks)
crew.kickoff()
private_mock.assert_called()

1954
tests/cassettes/test_agent_function_calling_llm.yaml Normal file
View File

File diff suppressed because it is too large Load Diff

2315
tests/cassettes/test_crew_function_calling_llm.yaml Normal file
View File

File diff suppressed because it is too large Load Diff

35
tests/crew_test.py
View File

@@ -492,3 +492,38 @@ def test_dont_set_agents_step_callback_if_already_set():
crew.kickoff()
assert researcher_agent.step_callback is not crew_callback
assert researcher_agent.step_callback is agent_callback
@pytest.mark.vcr(filter_headers=["authorization"])
def test_crew_function_calling_llm():
from unittest.mock import patch
from langchain.tools import tool
from langchain_openai import ChatOpenAI
llm = ChatOpenAI(model="gpt-3.5")
with patch.object(llm.client, "create", wraps=llm.client.create) as private_mock:
@tool
def learn_about_AI(topic) -> float:
"""Useful for when you need to learn about AI to write an paragraph about it."""
return "AI is a very broad field."
agent1 = Agent(
role="test role",
goal="test goal",
backstory="test backstory",
tools=[learn_about_AI],
)
essay = Task(
description="Write and then review an small paragraph on AI until it's AMAZING",
agent=agent1,
)
tasks = [essay]
print(agent1.function_calling_llm)
crew = Crew(agents=[agent1], tasks=tasks, function_calling_llm=llm)
print(agent1.function_calling_llm)
crew.kickoff()
private_mock.assert_called()