Can I add Fewshot tempalte/Fewshot examples to prebuilt react agent?

[mraguth]

This is my simple React agent and I like to add few shot examples along with a prompt in this example. Is that possible? Please help.

# First we initialize the model we want to use.
from langchain_openai import ChatOpenAI

model = llm


# For this tutorial we will use custom tool that returns pre-defined values for weather in two cities (NYC & SF)

from typing import Literal

from langchain_core.tools import tool


@tool
def get_weather(city: Literal["nyc", "sf"]):
    """Use this to get weather information."""
    if city == "nyc":
        return "It might be cloudy in nyc"
    elif city == "sf":
        return "It's always sunny in sf"
    else:
        raise AssertionError("Unknown city")


tools = [get_weather]

# We can add our system prompt here

prompt = "Respond in Hindi"

# Define the graph

from langgraph.prebuilt import create_react_agent

graph = create_react_agent(model, tools=tools, state_modifier=prompt)

def print_stream(stream):
    for s in stream:
        message = s["messages"][-1]
        if isinstance(message, tuple):
            print(message)
        else:
            message.pretty_print()

inputs = {"messages": [("user", "What's the weather in NYC?")]}

print_stream(graph.stream(inputs, stream_mode="values"))

[dosubot[bot]]

To add few-shot examples to a prebuilt React agent in LangChain, you can use the FewShotPromptTemplate or FewShotChatMessagePromptTemplate classes. These classes allow you to provide examples of desired interactions, which the language model can use to generate responses that align with your expectations.

Here's a general approach you can follow:

1. Create Few-Shot Examples: Define a few-shot prompt with examples of interactions between the user and the bot. These examples should demonstrate the type of responses you expect from the model.

2. Integrate Few-Shot Prompt: Use the FewShotPromptTemplate or FewShotChatMessagePromptTemplate to integrate these examples into the final prompt for the agent.

Unfortunately, I don't have the exact code snippet for integrating this into your specific setup, but you can refer to the LangChain documentation or examples for guidance on using these classes.

For more details, you can check a similar issue that was discussed and resolved here ^[1].

To continue talking to Dosu, mention @dosu.

[mraguth]

@dosu Assume that I created I created a few_shot_prompt. Now, how I inject that in our langgraph react agent? can you help?

few_shot_prompt = FewShotChatMessagePromptTemplate(
    examples=examples,
    example_prompt=example_prompt,
)

React agent is below and I like to inject 'few_shot_promt'

# Register tools and create the agent
tools = [tc_skills_v]

from langgraph.prebuilt import create_react_agent
graph = create_react_agent(llm, tools=tools)

from IPython.display import Image, display

#display(Image(graph.get_graph().draw_mermaid_png()))

def print_stream(stream):
    for s in stream:
        message = s["messages"][-1]
        if isinstance(message, tuple):
            print(message)
        else:
            message.pretty_print()


inputs = {"messages": [("user", "Write what employer is asking for")]}
print_stream(graph.stream(inputs, stream_mode="values"))

[dosubot[bot]]

To inject the few_shot_prompt into your LangChain React agent, you need to incorporate it into the prompt configuration of your agent. Here's how you can do it:

1. Modify the Prompt: Use the FewShotChatMessagePromptTemplate as part of the prompt setup for your agent. You can integrate it into the create_react_agent function by including it in the prompt configuration.

2. Update the Agent Creation: When creating the agent, pass the few_shot_prompt as part of the prompt or state modifier.

Here's an example of how you might modify your code:

# Assuming you have already created your few_shot_prompt
few_shot_prompt = FewShotChatMessagePromptTemplate(
    examples=examples,
    example_prompt=example_prompt,
)

# Register tools and create the agent
tools = [tc_skills_v]

# Integrate the few_shot_prompt into the agent
from langgraph.prebuilt import create_react_agent
graph = create_react_agent(llm, tools=tools, state_modifier=few_shot_prompt)

def print_stream(stream):
    for s in stream:
        message = s["messages"][-1]
        if isinstance(message, tuple):
            print(message)
        else:
            message.pretty_print()

inputs = {"messages": [("user", "Write what employer is asking for")]}
print_stream(graph.stream(inputs, stream_mode="values"))

In this setup, the few_shot_prompt is used as the state_modifier when creating the agent, which allows the few-shot examples to influence the agent's responses. Adjust the integration based on how your create_react_agent function is designed to accept and use prompts.

To continue talking to Dosu, mention @dosu.

[mraguth]

@dosu... in the above code, how to write my output to a file called 'output.txt' and want to store in the 'c:/' directory. Can you help?

[dosubot[bot]]

To write the output to a file called output.txt in the C:/ directory, you can modify the print_stream function to write the messages to a file instead of printing them. Here's how you can do it:

def write_stream_to_file(stream, file_path):
    with open(file_path, 'w', encoding='utf-8') as file:
        for s in stream:
            message = s["messages"][-1]
            if isinstance(message, tuple):
                file.write(str(message) + '\n')
            else:
                file.write(message.pretty_print() + '\n')

# Specify the file path
file_path = 'C:/output.txt'

inputs = {"messages": [("user", "Write what employer is asking for")]}
write_stream_to_file(graph.stream(inputs, stream_mode="values"), file_path)

This function opens output.txt in write mode and writes each message from the stream to the file. Make sure you have the necessary permissions to write to the C:/ directory. Adjust the file path if needed.

To continue talking to Dosu, mention @dosu.

[mraguth]

@dosu Error

{
	"name": "TypeError",
	"message": "unsupported operand type(s) for +: 'NoneType' and 'str'",
	"stack": "---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Cell In[49], line 83
     81 inputs = {\"messages\": [(\"user\", \"Write what employer is asking for\")]}
     82 #print_stream(graph.stream(inputs, stream_mode=\"values\"))
---> 83 write_stream_to_file(graph.stream(inputs, stream_mode=\"values\"), file_path)

Cell In[49], line 76, in write_stream_to_file(stream, file_path)
     74     file.write(str(message) + '\
')
     75 else:
---> 76     file.write(message.pretty_print() + '\
')

TypeError: unsupported operand type(s) for +: 'NoneType' and 'str'"
}

[dosubot[bot]]

Could you provide more details about the message.pretty_print() function? It seems like it might be returning None, which is causing the error when trying to concatenate it with a string. If pretty_print() is supposed to return a string, you might need to check its implementation or ensure that it returns a valid string. Alternatively, you can try converting the message to a string using str(message) if message itself is not None.

To continue talking to Dosu, mention @dosu.

[mraguth]

@dosu....instead of streaming the file, simpy create a output and direct the output to txt file? is that possible?

I don't need streaming here

def write_stream_to_file(stream, file_path):
with open(file_path, 'w', encoding='utf-8') as file:
for s in stream:
message = s["messages"][-1]
if isinstance(message, tuple):
file.write(str(message) + '\n')
else:
file.write(message.pretty_print() + '\n')

Specify the file path

file_path = 'C:/ai_req.txt'

inputs = {"messages": [("user", "Write what employer is asking for")]}
#print_stream(graph.stream(inputs, stream_mode="values"))
write_stream_to_file(graph.stream(inputs, stream_mode="values"), file_path)

[mraguth]

@dosu My above query stuck in endless loop and getting this error. How I mitigate?

GraphRecursionError: Recursion limit of 25 reached without hitting a stop condition. You can increase the limit by setting the recursion_limit config key.

[dosubot[bot]]

To avoid the endless loop and the GraphRecursionError, you can try the following steps:

1. Check for Recursion in Your Graph: Ensure that your graph does not have any recursive calls that could lead to an infinite loop. This might involve reviewing the logic in your create_react_agent setup and any tools or prompts that might be causing repeated calls.

2. Increase the Recursion Limit: If the recursion is intentional and necessary, you can increase the recursion limit by setting the recursion_limit configuration key. However, this should be done with caution, as it might mask underlying issues with your graph's logic.

3. Simplify the Graph: If possible, simplify the graph to reduce complexity and potential recursion paths. This might involve breaking down complex tasks into simpler, non-recursive components.

4. Debugging: Add logging or print statements to trace the execution flow and identify where the recursion is occurring. This can help pinpoint the exact part of the graph that is causing the issue.

If you want to simply generate the output and write it to a file without streaming, you can modify your approach to collect the output first and then write it to a file. Here's a basic example:

def generate_output(inputs):
    # Collect the output from the graph
    output = []
    for s in graph.stream(inputs, stream_mode="values"):
        message = s["messages"][-1]
        if isinstance(message, tuple):
            output.append(str(message))
        else:
            output.append(message.pretty_print())
    return output

def write_output_to_file(output, file_path):
    with open(file_path, 'w', encoding='utf-8') as file:
        for line in output:
            file.write(line + '\n')

# Specify the file path
file_path = 'C:/ai_req.txt'

inputs = {"messages": [("user", "Write what employer is asking for")]}
output = generate_output(inputs)
write_output_to_file(output, file_path)

This approach collects all the output first and then writes it to the file, avoiding the need for streaming.

To continue talking to Dosu, mention @dosu.

[mraguth]

@dosu.... but the job is running until it get error : GraphRecursionError: Recursion limit of 25 reached without hitting a stop condition. You can increase the limit by setting the recursion_limit config key.

[dosubot[bot]]

To address the GraphRecursionError you're encountering, you can try the following:

1. Increase the Recursion Limit: If the recursion is necessary, you can increase the recursion limit by setting the recursion_limit in the RunnableConfig. This can be done when invoking the graph. Here's an example:

   inputs = {"messages": [("user", "Write what employer is asking for")]}
   config = {'recursion_limit': 50}  # Increase the limit as needed
   output = graph.invoke(inputs, config=config)

2. Review the Graph Logic: Ensure that there are no unintended recursive calls in your graph setup. Check the logic in your create_react_agent function and any tools or prompts that might be causing repeated calls.

3. Debugging: Add logging or print statements to trace the execution flow and identify where the recursion is occurring. This can help pinpoint the exact part of the graph that is causing the issue.

By increasing the recursion limit, you allow more steps before hitting the limit, but it's important to ensure that the recursion is intentional and not due to a logical error in the graph setup.

To continue talking to Dosu, mention @dosu.

[sreenathelloti]

How can we integrate dynamic few shot prompting with langgraph react agent while implementing SQL agent