Output parser of langchain Agent

[KosPsych]

Checked other resources

• I added a very descriptive title to this question.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.

Commit to Help

• I commit to help with one of those options 👆

Example Code

class AgentParser(BaseLLMOutputParser):

    def __init__(self):
        super().__init__()
      
    
    def parse(self, text):
        print(text)
        return text
      

    def parse_result(self, output):
        print(output)
        return output

agent_output_parser=AgentOutputParser()
     self.mrkl = initialize_agent(
                tools, 
                llm,
                output_parser= agent_output_parser,
                agent_executor_kwargs={ "output_parser": agent_output_parser}
            )

Description

I am initializing a langchain agent as seen in the code.
However, using print I would expect to see the output of the agent parsed and somehow manipulate it.
In my current code I can't seem to handle the output of the agent and the prints i wrote never happen.
Is there an alternative way to pass a custom output parser?

System Info

langchain==0.1.0

[dosubot[bot]]

Hello @KosPsych! 👋 I'm Dosu, a bot here to assist you with your LangChain queries, debugging dilemmas, and to guide you on your way to becoming a contributor. While we await a human maintainer, I'm here to provide initial support and answer any questions you might have. Let's dive into your issue and see how we can resolve it together!

[KosPsych]

@dosu-bot Answer the question I posted

[dosubot[bot]]

Based on the code you've provided, it seems like you're trying to use a custom output parser with the initialize_agent function in LangChain version 0.1.0. However, there seems to be a mismatch between the class you've defined (AgentParser) and the instance you're trying to create (agent_output_parser=AgentOutputParser()).

You should be creating an instance of AgentParser instead of AgentOutputParser. Here's how you can do it:

agent_output_parser = AgentParser()

Also, it's important to note that the initialize_agent function in LangChain version 0.1.0 doesn't seem to directly accept an output_parser argument based on the source code.

However, you're passing output_parser to agent_executor_kwargs which is not correct. The agent_executor_kwargs is used to pass arguments to the AgentExecutor class, not the Agent class. The AgentExecutor class does not use an output_parser.

Instead, you should pass the output_parser to the agent_kwargs which is used to pass arguments to the Agent class. Here's how you can do it:

self.mrkl = initialize_agent(
    tools, 
    llm,
    agent_kwargs={ "output_parser": agent_output_parser}
)

Please note that the Agent class should be designed to accept an output_parser in its constructor and use it to parse the output. If it's not designed this way, you might need to modify the Agent class or use a different approach.

Also, please note that the initialize_agent function is deprecated since version 0.1.0 and it's recommended to use new agent constructor methods like create_react_agent, create_json_agent, create_structured_chat_agent, etc. You might want to consider upgrading to a newer version of LangChain if possible.

I hope this helps! If you have any other questions or need further clarification, feel free to ask.

Sources

• Output parser of langchain React Agent
• libs/core/langchain_core/output_parsers/base.py
• libs/langchain/langchain/agents/initialize.py
• libs/langchain/langchain/agents/chat/output_parser.py
• libs/langchain/langchain/agents/react/output_parser.py

About Dosu

This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

If you want to continue the conversation, start your reply with @dosu-bot.

[KosPsych]

@dosu-bot The provided solution does not work. is there and alternative?

[dosubot[bot]]

Based on the information you've provided, it seems like you're trying to pass a custom output parser to the initialize_agent function in LangChain version 0.1.0. However, the approach you're using seems to be incorrect.

In LangChain version 0.1.0, the initialize_agent function does not accept agent_kwargs as a parameter. Instead, it directly accepts an output_parser parameter. Therefore, you should create an instance of your custom AgentOutputParser and pass it directly to the output_parser parameter of the create_react_agent function.

Here's how you can do it:

1. Define your custom output parser by extending AgentOutputParser.
2. Create an instance of your custom parser.
3. Pass this instance as the output_parser argument to create_react_agent.

Here's an example:

from langchain.agents import AgentOutputParser

# Step 1: Define your custom output parser
class MyCustomOutputParser(AgentOutputParser):
    def parse(self, input, output):
        # Implement your parsing logic here
        # For example, print the output
        print("Custom parser output:", output)
        return super().parse(input, output)

# Step 2: Create an instance of your custom parser
custom_parser = MyCustomOutputParser()

# When calling create_react_agent, pass your custom parser as the output_parser argument
agent = create_react_agent(model, tools, prompt, output_parser=custom_parser)

This way, your custom output parser will be used by the agent, and any print statements or other logic inside your parser's parse method will be executed.

You can find more information about this in the source code of the create_react_agent function.

I hope this helps! If you have any other questions, feel free to ask.

Sources

• Output parser of langchain React Agent
• libs/langchain/langchain/agents/react/agent.py
• libs/langchain/langchain/agents/initialize.py

About Dosu

This response is meant to be useful and save you time. It is not meant to be a precise solution, but rather a starting point for your own research.

Help me be more useful! Please leave a 👍 if this is helpful and 👎 if it is irrelevant.

If you want to continue the conversation, start your reply with @dosu-bot.

[sumansaurav-talentica]

@dosu-bot I am creating an agent like agent = create_openai_tools_agent(self.model, self.tools, self.prompt)
I want the final response from agent in a formatted manner like below class

class Response(BaseModel):
"""Final response to the users query"""

answer: str = Field(description="The final answer to respond to the user")
components: List[SanitizedComponents] = Field(
    description="Modified form components. Return this even if there is not modification"
)

please suggest me the same code