feat: add message_output and refactor LCAgentComponent (#2755)
* feat(agent.py): add support for handling message responses in LCAgentComponent to improve agent functionality and interaction with messages * feat: add ToolEnabledLanguageModel type alias to constants.py This commit adds a new type alias `ToolEnabledLanguageModel` to the `constants.py` file in the `field_typing` module. This type alias is used to define the type of language models that have tooling enabled. It includes the `BaseLanguageModel`, `BaseLLM`, and `BaseChatModel` types. This change enhances the typing capabilities of the codebase and improves the clarity of the code. * feat: update agent.py to include support for ToolEnabledLanguageModel This commit modifies the `agent.py` file to include support for the `ToolEnabledLanguageModel` in the `LCToolsAgentComponent` class. The `ToolEnabledLanguageModel` is now added as an input type in the `HandleInput` section. This change enhances the functionality of the agent by allowing it to work with language models that have tooling enabled. The addition of this input type improves the flexibility and versatility of the codebase. * feat: add support for ToolEnabledLanguageModel in LCAgentComponent This commit modifies the `agent.py` file to include support for the `ToolEnabledLanguageModel` in the `LCAgentComponent` class. The `ToolEnabledLanguageModel` is now added as an input type in the `HandleInput` section. This change enhances the functionality of the agent by allowing it to work with language models that have tooling enabled. The addition of this input type improves the flexibility and versatility of the codebase. * feat: add AgentAsyncHandler for handling callbacks from Agents This commit adds the `AgentAsyncHandler` class to handle callbacks from langchain in the `callback.py` file. The `AgentAsyncHandler` is an async callback handler that can be used to handle various events such as tool start, tool end, agent action, and agent finish. This change enhances the functionality of the codebase by providing a convenient way to handle callbacks from langchain and log the events if a log function is provided. * chore: add field_serializer decorator to Log class for message serialization This commit adds the `field_serializer` decorator to the `Log` class in the `schema.py` file. The decorator is used to serialize the `message` attribute of the `Log` class, ensuring that all nested objects are properly serialized. This change improves the serialization process and enhances the functionality of the codebase. * feat: Fix issue with logs in LangSmithTracer This commit fixes an issue in the `LangSmithTracer` class where logs were not being properly serialized. The `add_metadata` method now converts logs to dictionaries using the `model_dump` method if they are not already dictionaries. This ensures that all logs are correctly serialized and improves the functionality of the codebase. * feat: Add support for Pydantic V1 models in Log serialization * fix: Update LCAgentComponent to handle list results in result variable This commit modifies the `LCAgentComponent` class in the `agent.py` file to handle list results in the `result` variable. If the `result` is a list, it is joined into a single string using the `text` attribute of each result dictionary. This change improves the functionality of the codebase by ensuring that the `result` variable is always a string, which is expected by the `Message` class. * feat: Add "name" parameter to AgentAsyncHandler methods This commit adds a "name" parameter to the `on_tool_start`, `on_tool_end`, `on_agent_action`, and `on_agent_finish` methods of the `AgentAsyncHandler` class in the `callback.py` file. The "name" parameter allows for specifying a name for the event, which can be useful for logging and tracking purposes. This change enhances the functionality of the codebase by providing more flexibility in handling callbacks and improves the clarity of the code. * feat: Update AgentAsyncHandler constructor to support logging multiple loggable types This commit updates the `AgentAsyncHandler` constructor in the `callback.py` file to support logging multiple loggable types. The `log_function` parameter now accepts a callable that can handle either a single `LoggableType` or a list of `LoggableType` objects along with a string representing the event name. This change enhances the flexibility of the codebase by allowing for more versatile logging options and improves the clarity of the code. * refactor(callback.py): update type hints in AgentAsyncHandler constructor and methods for better readability and accuracy refactor(callback.py): simplify on_tool_end method by using **kwargs for flexibility and consistency with other methods
This commit is contained in:
Gabriel Luiz Freitas Almeida
GitHub
parent
d4f5d149a8
commit
2c3e93bd87
5 changed files with 187 additions and 9 deletions
|
|
@ -1,20 +1,26 @@
|
|||
from abc import abstractmethod
|
||||
from typing import List
|
||||
from typing import List, Optional, Union, cast
|
||||
|
||||
from langchain.agents import AgentExecutor, BaseMultiActionAgent, BaseSingleActionAgent
|
||||
from langchain.agents.agent import RunnableAgent
|
||||
|
||||
from langchain.agents import AgentExecutor
|
||||
from langchain_core.messages import BaseMessage
|
||||
from langchain_core.runnables import Runnable
|
||||
|
||||
from langflow.base.agents.callback import AgentAsyncHandler
|
||||
from langflow.base.agents.utils import data_to_messages
|
||||
from langflow.custom import Component
|
||||
from langflow.inputs import BoolInput, IntInput, HandleInput
|
||||
from langflow.inputs.inputs import InputTypes
|
||||
from langflow.field_typing import Text, Tool
|
||||
from langflow.inputs.inputs import DataInput, InputTypes
|
||||
from langflow.io import BoolInput, HandleInput, IntInput, MessageTextInput
|
||||
from langflow.schema import Data
|
||||
from langflow.schema.message import Message
|
||||
from langflow.template import Output
|
||||
|
||||
|
||||
class LCAgentComponent(Component):
|
||||
trace_type = "agent"
|
||||
_base_inputs: List[InputTypes] = [
|
||||
MessageTextInput(name="input_value", display_name="Input"),
|
||||
BoolInput(
|
||||
name="handle_parsing_errors",
|
||||
display_name="Handle Parse Errors",
|
||||
|
|
@ -37,8 +43,24 @@ class LCAgentComponent(Component):
|
|||
|
||||
outputs = [
|
||||
Output(display_name="Agent", name="agent", method="build_agent"),
|
||||
Output(display_name="Response", name="response", method="message_response"),
|
||||
]
|
||||
|
||||
async def message_response(self) -> Message:
|
||||
agent = self.build_agent()
|
||||
result = await self.run_agent(
|
||||
agent=agent,
|
||||
inputs=self.input_value,
|
||||
tools=self.tools,
|
||||
message_history=self.chat_history,
|
||||
handle_parsing_errors=self.handle_parsing_errors,
|
||||
)
|
||||
if isinstance(result, list):
|
||||
result = "\n".join([result_dict["text"] for result_dict in result])
|
||||
message = Message(text=result, sender="Machine")
|
||||
self.status = message
|
||||
return message
|
||||
|
||||
def _validate_outputs(self):
|
||||
required_output_methods = ["build_agent"]
|
||||
output_names = [output.name for output in self.outputs]
|
||||
|
|
@ -65,6 +87,33 @@ class LCAgentComponent(Component):
|
|||
}
|
||||
return {**base, "agent_executor_kwargs": agent_kwargs}
|
||||
|
||||
async def run_agent(
|
||||
self,
|
||||
agent: Union[Runnable, BaseSingleActionAgent, BaseMultiActionAgent, AgentExecutor],
|
||||
inputs: str,
|
||||
tools: List[Tool],
|
||||
message_history: Optional[List[Data]] = None,
|
||||
handle_parsing_errors: bool = True,
|
||||
) -> Text:
|
||||
if isinstance(agent, AgentExecutor):
|
||||
runnable = agent
|
||||
else:
|
||||
runnable = AgentExecutor.from_agent_and_tools(
|
||||
agent=agent, # type: ignore
|
||||
tools=tools,
|
||||
verbose=True,
|
||||
handle_parsing_errors=handle_parsing_errors,
|
||||
)
|
||||
input_dict: dict[str, str | list[BaseMessage]] = {"input": inputs}
|
||||
if message_history:
|
||||
input_dict["chat_history"] = data_to_messages(message_history)
|
||||
result = await runnable.ainvoke(input_dict, config={"callbacks": [AgentAsyncHandler(self.log)]})
|
||||
self.status = result
|
||||
if "output" not in result:
|
||||
raise ValueError("Output key not found in result. Tried 'output'.")
|
||||
|
||||
return cast(str, result.get("output"))
|
||||
|
||||
|
||||
class LCToolsAgentComponent(LCAgentComponent):
|
||||
_base_inputs = LCAgentComponent._base_inputs + [
|
||||
|
|
@ -74,7 +123,13 @@ class LCToolsAgentComponent(LCAgentComponent):
|
|||
input_types=["Tool", "BaseTool"],
|
||||
is_list=True,
|
||||
),
|
||||
HandleInput(name="llm", display_name="Language Model", input_types=["LanguageModel"], required=True),
|
||||
HandleInput(
|
||||
name="llm",
|
||||
display_name="Language Model",
|
||||
input_types=["LanguageModel", "ToolEnabledLanguageModel"],
|
||||
required=True,
|
||||
),
|
||||
DataInput(name="chat_history", display_name="Chat History", is_list=True),
|
||||
]
|
||||
|
||||
def build_agent(self) -> AgentExecutor:
|
||||
|
|
|
|||
103
src/backend/base/langflow/base/agents/callback.py
Normal file
103
src/backend/base/langflow/base/agents/callback.py
Normal file
|
|
@ -0,0 +1,103 @@
|
|||
from typing import Any, Callable, Concatenate, Dict, List
|
||||
from uuid import UUID
|
||||
|
||||
from langchain.callbacks.base import AsyncCallbackHandler
|
||||
from langchain_core.agents import AgentAction, AgentFinish
|
||||
|
||||
from langflow.schema.log import LoggableType
|
||||
|
||||
|
||||
class AgentAsyncHandler(AsyncCallbackHandler):
|
||||
"""Async callback handler that can be used to handle callbacks from langchain."""
|
||||
|
||||
def __init__(self, log_function: Callable[Concatenate[LoggableType | list[LoggableType], ...], None] | None = None):
|
||||
self.log_function = log_function
|
||||
|
||||
async def on_tool_start(
|
||||
self,
|
||||
serialized: Dict[str, Any],
|
||||
input_str: str,
|
||||
*,
|
||||
run_id: UUID,
|
||||
parent_run_id: UUID | None = None,
|
||||
tags: List[str] | None = None,
|
||||
metadata: Dict[str, Any] | None = None,
|
||||
inputs: Dict[str, Any] | None = None,
|
||||
**kwargs: Any,
|
||||
) -> None:
|
||||
if self.log_function is None:
|
||||
return
|
||||
self.log_function(
|
||||
{
|
||||
"type": "tool_start",
|
||||
"serialized": serialized,
|
||||
"input_str": input_str,
|
||||
"run_id": run_id,
|
||||
"parent_run_id": parent_run_id,
|
||||
"tags": tags,
|
||||
"metadata": metadata,
|
||||
"inputs": inputs,
|
||||
**kwargs,
|
||||
},
|
||||
name="Tool Start",
|
||||
)
|
||||
|
||||
async def on_tool_end(self, output: Any, *, run_id: UUID, parent_run_id: UUID | None = None, **kwargs: Any) -> None:
|
||||
if self.log_function is None:
|
||||
return
|
||||
self.log_function(
|
||||
{
|
||||
"type": "tool_end",
|
||||
"output": output,
|
||||
"run_id": run_id,
|
||||
"parent_run_id": parent_run_id,
|
||||
**kwargs,
|
||||
},
|
||||
name="Tool End",
|
||||
)
|
||||
|
||||
async def on_agent_action(
|
||||
self,
|
||||
action: AgentAction,
|
||||
*,
|
||||
run_id: UUID,
|
||||
parent_run_id: UUID | None = None,
|
||||
tags: List[str] | None = None,
|
||||
**kwargs: Any,
|
||||
) -> None:
|
||||
if self.log_function is None:
|
||||
return
|
||||
self.log_function(
|
||||
{
|
||||
"type": "agent_action",
|
||||
"action": action,
|
||||
"run_id": run_id,
|
||||
"parent_run_id": parent_run_id,
|
||||
"tags": tags,
|
||||
**kwargs,
|
||||
},
|
||||
name="Agent Action",
|
||||
)
|
||||
|
||||
async def on_agent_finish(
|
||||
self,
|
||||
finish: AgentFinish,
|
||||
*,
|
||||
run_id: UUID,
|
||||
parent_run_id: UUID | None = None,
|
||||
tags: List[str] | None = None,
|
||||
**kwargs: Any,
|
||||
) -> None:
|
||||
if self.log_function is None:
|
||||
return
|
||||
self.log_function(
|
||||
{
|
||||
"type": "agent_finish",
|
||||
"finish": finish,
|
||||
"run_id": run_id,
|
||||
"parent_run_id": parent_run_id,
|
||||
"tags": tags,
|
||||
**kwargs,
|
||||
},
|
||||
name="Agent Finish",
|
||||
)
|
||||
|
|
@ -21,6 +21,7 @@ from langflow.schema.message import Message
|
|||
|
||||
NestedDict: TypeAlias = Dict[str, Union[str, Dict]]
|
||||
LanguageModel = TypeVar("LanguageModel", BaseLanguageModel, BaseLLM, BaseChatModel)
|
||||
ToolEnabledLanguageModel = TypeVar("ToolEnabledLanguageModel", BaseLanguageModel, BaseLLM, BaseChatModel)
|
||||
Retriever = TypeVar(
|
||||
"Retriever",
|
||||
BaseRetriever,
|
||||
|
|
|
|||
|
|
@ -120,7 +120,8 @@ class LangSmithTracer(BaseTracer):
|
|||
raw_outputs = outputs
|
||||
processed_outputs = self._convert_to_langchain_types(outputs)
|
||||
if logs:
|
||||
child.add_metadata(self._convert_to_langchain_types({"logs": {log.get("name"): log for log in logs}}))
|
||||
logs_dicts = [log if isinstance(log, dict) else log.model_dump() for log in logs]
|
||||
child.add_metadata(self._convert_to_langchain_types({"logs": {log.get("name"): log for log in logs_dicts}}))
|
||||
child.add_metadata(self._convert_to_langchain_types({"outputs": raw_outputs}))
|
||||
child.end(outputs=processed_outputs, error=self._error_to_string(error))
|
||||
if error:
|
||||
|
|
|
|||
|
|
@ -1,9 +1,27 @@
|
|||
from typing_extensions import TypedDict
|
||||
from pydantic import BaseModel, field_serializer
|
||||
from pydantic.v1 import BaseModel as V1BaseModel
|
||||
|
||||
from langflow.schema.log import LoggableType
|
||||
|
||||
|
||||
class Log(TypedDict):
|
||||
class Log(BaseModel):
|
||||
name: str
|
||||
message: LoggableType
|
||||
type: str
|
||||
|
||||
@field_serializer("message")
|
||||
def serialize_message(self, value):
|
||||
# We need to make sure everything inside the message has been serialized
|
||||
if isinstance(value, dict):
|
||||
return {key: self.serialize_message(value[key]) for key in value}
|
||||
if isinstance(value, list):
|
||||
return [self.serialize_message(item) for item in value]
|
||||
# To json is for LangChain Serializable objects
|
||||
if hasattr(value, "dict") and isinstance(value, V1BaseModel):
|
||||
# This is for Pydantic V1 models
|
||||
return value.dict()
|
||||
if hasattr(value, "to_json"):
|
||||
return value.to_json()
|
||||
if isinstance(value, BaseModel):
|
||||
return value.model_dump(exclude_none=True)
|
||||
return value
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue