Brokers in LangChain allow extra dynamic and versatile workflows by permitting fashions to work together with numerous exterior instruments, APIs, plugins or different fashions based mostly on the context of the duty at hand. Some examples of those embrace connecting to Wikipedia, Google search, Python, Twilio or Expedia API.
The code instance beneath builds an agent with the wikipedia API.
from langchain.brokers import create_tool_calling_agent, load_tools, initialize_agent, AgentType
%pip set up --upgrade --quiet wikipedia
instruments = load_tools(["arxiv", "wikipedia"], llm=llm)
The variable Verbose = True particulars the inner state of the agent object highlighting which instruments are used within the chain.
agent = initialize_agent(
instruments,
llm,
agent = AgentType.ZERO_SHOT_REACT_DESCRIPTION,
verbose = True
)agent.run("Who's the present chief of Afghanistan")
Reminiscence in Langchain extends dialog reminiscence, empowering LLM based mostly functions like chatbots to take care of and leverage context over a number of interactions.
Take into account the sooner chain created on this put up. Working the rain with two completely different prompts and checking the reminiscence with chain.reminiscence outputs None.
prompt_template_for_vacation = PromptTemplate(
input_variables = ['continent'],
template = "I am planning a vaction.Recommend one nation to go to in {continent}. Give solely the nation identify"
)
chain = LLMChain (llm = llm, immediate=prompt_template_for_vacation)nation = chain.run("America")
to_markdown(nation)
chain = LLMChain (llm = llm, immediate=prompt_template_for_vacation)nation = chain.run("Africa")
to_markdown(nation)
print(chain.reminiscence)
Dialog Buffer Reminiscence
Nevertheless reminiscence may be added utilizing ConversationBufferMemory
from langchain.reminiscence import ConversationBufferMemoryreminiscence = ConversationBufferMemory()
Working two new prompts after instantiating reminiscence would return a reminiscence historical past when chain.reminiscence is computed this time round.
chain = LLMChain (llm = llm, immediate=prompt_template_for_vacation, reminiscence=reminiscence)nation = chain.run("Europe")
to_markdown(nation)
nation = chain.run("Asia")to_markdown(nation)
print(chain.reminiscence.buffer)
Although ConversationBufferMemory creates reminiscence, it additionally sends all content material in its historical past to the LLM — this would possibly use extra tokens and incurr extra prices.
Dialog Chain Reminiscence
A reminiscence window permits the consumer to decide on how far again prompts are saved. Implementing ConversationChain permits for a reminiscence window.
from langchain.chains import ConversationChainhistorical past = ConversationChain(llm=llm)
print(historical past.immediate.template)
historical past.run("what date is the subsequent US presidential elections? ")
print(historical past.reminiscence.buffer)
Dialog Buffer Window Reminiscence
Within the code pattern beneath, a reminiscence window of okay = 1 is outlined i.e the reminiscence would solely retailer the final immediate
from langchain.reminiscence import ConversationBufferWindowMemoryreminiscence = ConversationBufferWindowMemory(okay=1)
historical past = ConversationChain(llm=llm,reminiscence=reminiscence)
historical past.run("what date is the subsequent US presidential elections? ")
historical past.run("who's the richest girl on the earth")
The historical past.reminiscence would print the final immediate historical past for “who’s the richest girl on the earth” on this case
print(historical past.reminiscence.buffer)
