Conversational Agents

Overview

Conversational agents in LangChain are designed to have natural conversations with users, unlike traditional goal-oriented agents. This is accomplished using the conversational-react-description agent type along with a memory component.

Initializing the Conversational Agent

To create a conversational agent, you need to:

1. Define tools like search that the agent can use
2. Initialize a memory component
3. Initialize the agent by passing the tools, LLM, and memory

from langchain.agents import Tool  
from langchain.agents import AgentType
from langchain.memory import ConversationBufferMemory
from langchain import OpenAI
from langchain.utilities import SerpAPIWrapper
from langchain.agents import initialize_agent

search = SerpAPIWrapper()
tools = [
  Tool(
    name = "Current Search",
    func=search.run,
    description="useful for when you need to answer questions about current events or the current state of the world"
  ),
]

memory = ConversationBufferMemory(memory_key="chat_history") 

llm=OpenAI(temperature=0)
agent_chain = initialize_agent(tools, llm, agent=AgentType.CONVERSATIONAL_REACT_DESCRIPTION, verbose=True, memory=memory)

Using Memory

The ConversationBufferMemory tracks chat history and context.

from langchain.prompts import MessagesPlaceholder

chat_history = MessagesPlaceholder(variable_name="chat_history")

memory = ConversationBufferMemory(
  memory_key="chat_history", 
  return_messages=True,
  prompt=chat_history
)

You can customize the memory prompt to control how chat history is formatted.

Training Conversational Models

Fine-tuning on conversational data can improve coherence.

from langchain.llms import OpenAI
from langchain.chains import ConversationChain

llm = OpenAI(temperature=0)
chain = ConversationChain(llm=llm)

conv_dataset = [
  ("hi", "hello there!"),
  ("what's your name?", "I'm Clara, nice to meet you!"),
  ("how are you today?", "I'm doing great, thanks for asking!") 
]

chain.fine_tune(conv_dataset, batch_size=8, epochs=20)

Use a low learning rate like 1e-6 to gently fine-tune the model.

Troubleshooting

If you see repetitive or inconsistent responses:

• Lower the temperature to increase determinism
• Adjust the memory prompt formatting
• Fine-tune on more conversational data
• Try different prompt engineering strategies

Conclusion

With the right setup of tools, memory, and training, conversational agents in LangChain can enable natural dialog. Monitor logs and experiment with customization to improve performance.