How to get structured output out of a ReAct Agent

[invalidexplorer]

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

from langchain.output_parsers import ResponseSchema,StructuredOutputParser

code=ResponseSchema(name="code",description="well formatted python code")
description=ResponseSchema(name="description",description="approach of how the problem was solved, any errors or blockers if any faced")
output_parser=StructuredOutputParser.from_response_schemas([code,description])

from langchain_experimental.tools import PythonREPLTool
from langchain.agents import AgentExecutor, create_react_agent
from langchain_aws import ChatBedrockConverse

llm = ChatBedrockConverse(
    model="anthropic.claude-3-sonnet-20240229-v1:0",
    temperature=0,
    max_tokens=4096,
)


tools = [PythonREPLTool()]
agent = create_react_agent(llm, tools, prompt, output_parser=output_parser)
agent_executor = AgentExecutor(agent=agent, tools=tools, verbose=True, handle_parsing_errors=True, max_iterations=100)

agent_executor.invoke({"input": f'''Find last common ancenstor for 2 binary search trees
'''})

Description

I have tried to use ReAct agents and format their output using "from langchain.output_parsers import ResponseSchema,StructuredOutputParser

code=ResponseSchema(name="code",description="well formatted python code")
description=ResponseSchema(name="description",description="approach of how the problem was solved, any errors or blockers if any faced")
output_parser=StructuredOutputParser.from_response_schemas([code,description])" however the react agents were stuck on a loop "Got invalid JSON object. Error: Expecting value: line 1 column 1 (char 0)Invalid or incomplete responseGot invalid JSON object. Error: Expecting value: line 1 column 1 (char 0)Invalid or incomplete responseGot invalid JSON object. Error: Expecting value: line 1 column 1 (char 0)Invalid or incomplete response
". Would really appreciate on how to go about solving them.

System Info

System Information

OS: Darwin
OS Version: Darwin Kernel Version 23.5.0: Wed May 1 20:13:18 PDT 2024; root:xnu-10063.121.3~5/RELEASE_ARM64_T6030
Python Version: 3.12.4 | packaged by Anaconda, Inc. | (main, Jun 18 2024, 10:07:17) [Clang 14.0.6 ]

Package Information

langchain_core: 0.2.29
langchain: 0.2.6
langchain_community: 0.2.6
langsmith: 0.1.83
langchain_aws: 0.1.16
langchain_experimental: 0.0.62
langchain_openai: 0.1.21
langchain_text_splitters: 0.2.2
langchainhub: 0.1.20
langgraph: 0.2.3

Optional packages not installed

langserve

Other Dependencies

aiohttp: 3.9.5
async-timeout: Installed. No version info available.
boto3: 1.34.140
dataclasses-json: 0.6.7
jsonpatch: 1.33
langgraph-checkpoint: 1.0.2
numpy: 1.26.4
openai: 1.40.3
orjson: 3.10.6
packaging: 23.2
pydantic: 2.8.2
PyYAML: 6.0.1
requests: 2.32.2
SQLAlchemy: 2.0.30
tenacity: 8.2.2
tiktoken: 0.7.0
types-requests: 2.32.0.20240622
typing-extensions: 4.11.0

[dosubot[bot]]

Hello, @invalidexplorer! I'm here to assist you with any bugs, questions, or contributions you might have. Let's work together to resolve your issue.

To get structured output from a ReAct Agent in LangChain without encountering JSON parsing errors, you can use the ReActOutputParser class. This class is designed to handle ReAct-style LLM calls and ensures that the output is parsed correctly, whether it signals an action or a final answer.

Here is an example of how you can use it:

import re
from langchain_core.language_models import BaseLanguageModel
from langchain.agents.output_parsers.react import ReActOutputParser
from langchain_core.exceptions import OutputParserException

# Assuming you have an instance of a language model
llm = BaseLanguageModel()

# Create the parser
parser = ReActOutputParser()

# Example text to parse
text = """
Action: Final Answer[The result is 42]
"""

# Parse the text
try:
    result = parser.parse(text)
    print(result)
except OutputParserException as e:
    print(f"Error parsing output: {e}")

This approach ensures that the output is parsed correctly according to the expected format, reducing the likelihood of JSON parsing errors. The ReActOutputParser class uses a regex pattern to identify and extract the action and its input, and it handles both action and final answer outputs appropriately ^[1][2][3].

To continue talking to Dosu, mention @dosu.

[dosubot[bot]]

Hello, @invalidexplorer! I'm here to assist you with any bugs, questions, or contributions you might have. Let's work together to resolve your issue.

To get structured output from a ReAct Agent in LangChain without encountering JSON parsing errors, you can use the ReActOutputParser class. This class is designed to handle ReAct-style LLM calls and ensures that the output is parsed correctly, whether it signals an action or a final answer.

Here is an example of how you can use it:

import re
from langchain_core.language_models import BaseLanguageModel
from langchain.agents.output_parsers.react import ReActOutputParser
from langchain_core.exceptions import OutputParserException

# Assuming you have an instance of a language model
llm = BaseLanguageModel()

# Create the parser
parser = ReActOutputParser()

# Example text to parse
text = """
Action: Final Answer[The result is 42]
"""

# Parse the text
try:
    result = parser.parse(text)
    print(result)
except OutputParserException as e:
    print(f"Error parsing output: {e}")

This approach ensures that the output is parsed correctly according to the expected format, reducing the likelihood of JSON parsing errors. The ReActOutputParser class uses a regex pattern to identify and extract the action and its input, and it handles both action and final answer outputs appropriately ^[1][2][3].

To continue talking to Dosu, mention @dosu.

[invalidexplorer]

@dosu not what I am looking for. I am looking for the final output from reAct agent to follow a specific format.

[dosubot[bot]]

To ensure that the final output from a ReAct agent follows a specific format, you can use the ReActJsonSingleInputOutputParser class. This class is designed to parse outputs in a specific format and handle errors appropriately.

Here is how you can use it:

1. Define the Expected Output Format: Ensure that the output from the ReAct agent follows the expected format, typically enclosed in JSON within triple backticks:
   • For actions:

     ```json
     {
       "action": "search",
       "action_input": "what is the temperature in SF"
     }
     

   • For final answers:

     ```json
     {
       "action": "Final Answer",
       "action_input": "The temperature is 100 degrees"
     }
     

2. Use the ReActJsonSingleInputOutputParser Class: This class will parse the output and handle any parsing errors. If the output does not follow the expected format, it will raise an OutputParserException.

Here is an example of how to use the ReActJsonSingleInputOutputParser:

from langchain.agents.output_parsers.react_json_single_input import ReActJsonSingleInputOutputParser
from langchain_core.agents import AgentAction, AgentFinish
from langchain_core.exceptions import OutputParserException

# Initialize the parser
parser = ReActJsonSingleInputOutputParser()

# Example output from the ReAct agent
output_text = """
```json
{
  "action": "search",
  "action_input": "current temperature in SF"
}

"""

Parse the output

try:
result = parser.invoke(output_text)
if isinstance(result, AgentAction):
print("Parsed an action:", result)
elif isinstance(result, AgentFinish):
print("Parsed a final answer:", result)
except OutputParserException as e:
print("Error parsing output:", e)

By following these steps and using the `ReActJsonSingleInputOutputParser`, you can ensure that the final output from the ReAct agent follows the specific format you need and handle any parsing errors appropriately <sup>[[1]](https://github.com/langchain-ai/langchain/blob/master/libs/langchain/langchain/agents/output_parsers/react_json_single_input.py)</sup><sup>[[2]](https://github.com/langchain-ai/langchain/blob/master/libs/langchain/langchain/agents/react/output_parser.py)</sup><sup>[[3]](https://github.com/langchain-ai/langchain/blob/master/libs/langchain/langchain/agents/structured_chat/output_parser.py)</sup><sup>[[4]](https://github.com/langchain-ai/langchain/blob/master/libs/langchain/tests/unit_tests/agents/output_parsers/test_react_json_single_input.py)</sup>.


<!-- Dosu Comment Footer -->
*To continue talking to [Dosu](https://dosu.dev), mention @dosu.*

[Hesdi]

Hey, did you actually found a solution at the end?

[invalidexplorer]

Hey, did you actually found a solution at the end?

Not yet, I looked into the concept of validator nodes but I am looking for something that is compatible with langchain's react agents not langgraph's react agent.

[spatnala-eightfold]

Hey, did anyone find a solution for this ?