mirror of
https://github.com/langchain-ai/langgraph.git
synced 2026-08-26 01:22:24 +02:00
Sydney RunkleandGitHub
e269b46b1b
Change how we handle tool errors by default in tool node. 1. If a tool invocation failed due to arg validation, we return an artificial `ToolMessage` back to the model w/ a request for arg correction (the same, but we've improved the message) 2. If a tool execution failed for an unknown reason, we raise (different, we used to automatically return to model)
118 lines
3.8 KiB
Python
118 lines
3.8 KiB
Python
import json
|
|
from dataclasses import asdict, is_dataclass
|
|
from typing import (
|
|
Any,
|
|
Callable,
|
|
Dict,
|
|
Generic,
|
|
List,
|
|
Literal,
|
|
Optional,
|
|
Sequence,
|
|
Type,
|
|
Union,
|
|
)
|
|
|
|
from langchain_core.callbacks import CallbackManagerForLLMRun
|
|
from langchain_core.language_models import BaseChatModel, LanguageModelInput
|
|
from langchain_core.messages import (
|
|
AIMessage,
|
|
BaseMessage,
|
|
ToolCall,
|
|
)
|
|
from langchain_core.outputs import ChatGeneration, ChatResult
|
|
from langchain_core.runnables import Runnable
|
|
from langchain_core.tools import BaseTool
|
|
from pydantic import BaseModel
|
|
|
|
from langgraph.prebuilt.chat_agent_executor import StructuredResponseT
|
|
|
|
|
|
class FakeToolCallingModel(BaseChatModel, Generic[StructuredResponseT]):
|
|
tool_calls: Optional[Union[list[list[ToolCall]], list[list[dict]]]] = None
|
|
structured_response: Optional[StructuredResponseT] = None
|
|
index: int = 0
|
|
tool_style: Literal["openai", "anthropic"] = "openai"
|
|
|
|
def _generate(
|
|
self,
|
|
messages: List[BaseMessage],
|
|
stop: Optional[List[str]] = None,
|
|
run_manager: Optional[CallbackManagerForLLMRun] = None,
|
|
**kwargs: Any,
|
|
) -> ChatResult:
|
|
"""Top Level call"""
|
|
rf = kwargs.get("response_format")
|
|
is_native = isinstance(rf, dict) and rf.get("type") == "json_schema"
|
|
|
|
if self.tool_calls:
|
|
if is_native:
|
|
tool_calls = (
|
|
self.tool_calls[self.index]
|
|
if self.index < len(self.tool_calls)
|
|
else []
|
|
)
|
|
else:
|
|
tool_calls = self.tool_calls[self.index % len(self.tool_calls)]
|
|
else:
|
|
tool_calls = []
|
|
|
|
if is_native and not tool_calls:
|
|
if isinstance(self.structured_response, BaseModel):
|
|
content_obj = self.structured_response.model_dump()
|
|
elif is_dataclass(self.structured_response):
|
|
content_obj = asdict(self.structured_response)
|
|
elif isinstance(self.structured_response, dict):
|
|
content_obj = self.structured_response
|
|
message = AIMessage(content=json.dumps(content_obj), id=str(self.index))
|
|
else:
|
|
messages_string = "-".join([m.content for m in messages])
|
|
message = AIMessage(
|
|
content=messages_string,
|
|
id=str(self.index),
|
|
tool_calls=tool_calls.copy(),
|
|
)
|
|
self.index += 1
|
|
return ChatResult(generations=[ChatGeneration(message=message)])
|
|
|
|
@property
|
|
def _llm_type(self) -> str:
|
|
return "fake-tool-call-model"
|
|
|
|
def bind_tools(
|
|
self,
|
|
tools: Sequence[Union[Dict[str, Any], Type[BaseModel], Callable, BaseTool]],
|
|
**kwargs: Any,
|
|
) -> Runnable[LanguageModelInput, BaseMessage]:
|
|
if len(tools) == 0:
|
|
raise ValueError("Must provide at least one tool")
|
|
|
|
tool_dicts = []
|
|
for tool in tools:
|
|
if isinstance(tool, dict):
|
|
tool_dicts.append(tool)
|
|
continue
|
|
if not isinstance(tool, BaseTool):
|
|
raise TypeError(
|
|
"Only BaseTool and dict is supported by FakeToolCallingModel.bind_tools"
|
|
)
|
|
|
|
# NOTE: this is a simplified tool spec for testing purposes only
|
|
if self.tool_style == "openai":
|
|
tool_dicts.append(
|
|
{
|
|
"type": "function",
|
|
"function": {
|
|
"name": tool.name,
|
|
},
|
|
}
|
|
)
|
|
elif self.tool_style == "anthropic":
|
|
tool_dicts.append(
|
|
{
|
|
"name": tool.name,
|
|
}
|
|
)
|
|
|
|
return self.bind(tools=tool_dicts)
|