Langchain中Fake LLM调用PythonAstREPLTool报错,求正确实现方案
代码错误原因及修复
错误根源
错误源于工具名称不匹配:PythonAstREPLTool的默认工具名称是python_repl_ast,但你在FakeListLLM的响应里写的是Action: Python REPL,Agent无法识别这个未注册的工具名称,因此抛出错误。
修复方案(两种任选其一)
方案1:修改FakeLLM响应中的工具名称
将响应里的工具名称改为工具默认的python_repl_ast:
from langchain.agents import initialize_agent from langchain.llms.fake import FakeListLLM from langchain.agents import AgentType from langchain_experimental.tools import PythonAstREPLTool # 调整Action后的工具名称为默认值 res = ["Action: python_repl_ast\nAction Input: print(2.2 + 2.22)", "Final Answer: 4.42"] llm = FakeListLLM(responses=res) agent = initialize_agent(tools=[PythonAstREPLTool()], llm=llm, agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION, verbose=True ) agent.run("what is 2.2 + 2.22?")
方案2:给工具自定义名称
初始化PythonAstREPLTool时指定name参数为"Python REPL",和响应里的名称保持一致:
from langchain.agents import initialize_agent from langchain.llms.fake import FakeListLLM from langchain.agents import AgentType from langchain_experimental.tools import PythonAstREPLTool res = ["Action: Python REPL\nAction Input: print(2.2 + 2.22)", "Final Answer: 4.42"] llm = FakeListLLM(responses=res) # 给工具设置自定义名称 tool = PythonAstREPLTool(name="Python REPL") agent = initialize_agent(tools=[tool], llm=llm, agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION, verbose=True ) agent.run("what is 2.2 + 2.22?")
FakeListLLM的更优实现方式
FakeListLLM核心用于离线测试,无需调用真实LLM,以下是几种更适配测试场景的实现思路:
1. 按输入场景映射响应
针对多输入测试场景,自定义LLM类实现输入与响应的映射,替代固定顺序的列表:
from langchain.llms.base import LLM from typing import List, Optional class CustomFakeLLM(LLM): # 定义输入到响应的映射表 input_response_map: dict = {} @property def _llm_type(self) -> str: return "custom-fake" def _call(self, prompt: str, stop: Optional[List[str]] = None) -> str: # 匹配输入返回对应响应,无匹配则返回默认值 return self.input_response_map.get(prompt, "默认测试响应") # 使用示例 llm = CustomFakeLLM( input_response_map={ "what is 2.2 + 2.22?": "Action: python_repl_ast\nAction Input: print(2.2 + 2.22)", "what is 3*5?": "Action: python_repl_ast\nAction Input: print(3*5)", "hello": "Final Answer: Hello there!" } )
2. 模拟流式输出
若需要测试流式响应逻辑,自定义支持流式输出的Fake LLM:
from langchain.llms.base import LLM from typing import Iterator, List, Optional class StreamingFakeLLM(LLM): response: str = "" @property def _llm_type(self) -> str: return "streaming-fake" def _call(self, prompt: str, stop: Optional[List[str]] = None) -> str: return self.response def _stream(self, prompt: str, stop: Optional[List[str]] = None) -> Iterator[str]: # 逐字符返回,模拟流式输出效果 for char in self.response: yield char # 使用示例 llm = StreamingFakeLLM(response="Final Answer: 4.42") for chunk in llm.stream("what is 2.2 + 2.22?"): print(chunk, end="")
3. 动态生成响应模板
针对Agent的工具调用测试,预定义响应模板,根据输入参数动态填充内容,避免硬编码:
from langchain.llms.fake import FakeListLLM def get_calculation_responses(num1: float, num2: float, operator: str): result = eval(f"{num1}{operator}{num2}") return [ f"Action: python_repl_ast\nAction Input: print({num1}{operator}{num2})", f"Final Answer: {result}" ] # 使用示例 res = get_calculation_responses(2.2, 2.22, "+") llm = FakeListLLM(responses=res)
内容的提问来源于stack exchange,提问作者Himanshu Gulati
相关产品推荐
相关产品推荐

