LangChain学习之四种Memory模式使用

1. 学习背景

在LangChain for LLM应用程序开发中课程中，学习了LangChain框架扩展应用程序开发中语言模型的用例和功能的基本技能，遂做整理为后面的应用做准备。视频地址：基于LangChain的大语言模型应用开发+构建和评估。

2. 四种memory模式

本实验基于jupyternotebook进行。

2.1 ConversationBufferMemory

import warnings
warnings.filterwarnings('ignore')

from langchain.chat_models import ChatOpenAI
from langchain.chains import ConversationChain
from langchain.memory import ConversationBufferMemory


llm = ChatOpenAI(
    api_key = "XXXX",
    base_url = "XXXX",
    temperature=0.0
)

memory = ConversationBufferMemory()
conversation = ConversationChain(
    llm=llm, 
    memory = memory,
    verbose=True #设置为True可以看到模型的具体思考过程
)

conversation.predict(input="Hi, my name is Andrew")

输出如下：

> Entering new ConversationChain chain...
Prompt after formatting:
The following is a friendly conversation between a human and an AI. The AI is talkative and provides lots of specific details from its context. If the AI does not know the answer to a question, it truthfully says it does not know.

Current conversation:

Human: Hi, my name is Andrew
AI:

> Finished chain.
"Hello Andrew! It's nice to meet you. My name is AI and I'm here to assist you with any questions or tasks you have. How can I help you today?"

继续测试

conversation.predict(input="What is 1+1?")

输出如下：

> Entering new ConversationChain chain...
Prompt after formatting:
The following is a friendly conversation between a human and an AI. The AI is talkative and provides lots of specific details from its context. If the AI does not know the answer to a question, it truthfully says it does not know.

Current conversation:
Human: Hi, my name is Andrew
AI: Hello Andrew! It's nice to meet you. My name is AI and I'm here to assist you with any questions or tasks you have. How can I help you today?
Human: What is 1+1?
AI:

> Finished chain.
'1+1 equals 2. This is a basic mathematical operation that can be solved by adding the two numbers together. Is there anything else I can assist you with?'

再次输入之前的问题

conversation.predict(input="What is my name?")

输出如下：

> Entering new ConversationChain chain...
Prompt after formatting:
The following is a friendly conversation between a human and an AI. The AI is talkative and provides lots of specific details from its context. If the AI does not know the answer to a question, it truthfully says it does not know.

Current conversation:
Human: Hi, my name is Andrew
AI: Hello Andrew! It's nice to meet you. My name is AI and I'm here to assist you with any questions or tasks you have. How can I help you today?
Human: What is 1+1?
AI: 1+1 equals 2. This is a basic mathematical operation that can be solved by adding the two numbers together. Is there anything else I can assist you with?
Human: What is my name?
AI:

> Finished chain.
'Your name is Andrew.'

我们可以清楚的看到每次LLM所记忆的内容，看看buffer有什么

print(memory.buffer)

输出如下：

Human: Hi, my name is Andrew
AI: Hello Andrew! It's nice to meet you. My name is AI and I'm here to assist you with any questions or tasks you have. How can I help you today?
Human: What is 1+1?
AI: 1+1 equals 2. This is a basic mathematical operation that can be solved by adding the two numbers together. Is there anything else I can assist you with?
Human: What is my name?
AI: Your name is Andrew.

memory.load_memory_variables({})

输出如下：

{'history': "Human: Hi, my name is Andrew\nAI: Hello Andrew! It's nice to meet you. My name is AI and I'm here to assist you with any questions or tasks you have. How can I help you today?\nHuman: What is 1+1?\nAI: 1+1 equals 2. This is a basic mathematical operation that can be solved by adding the two numbers together. Is there anything else I can assist you with?\nHuman: What is my name?\nAI: Your name is Andrew."}

接下来，我们尝试一下对ConversationBufferMemory的操作。

# 重新初始化memory
memory = ConversationBufferMemory()
memory.save_context({"input": "Hi"}, 
                    {"output": "What's up"})
print(memory.buffer)

输出如下：

Human: Hi
AI: What's up

以变量形式读取

memory.load_memory_variables({})

输出如下：

{'history': "Human: Hi\nAI: What's up"}

再往里存入内容

memory.save_context({"input": "Not much, just hanging"}, 
                    {"output": "Cool"})
memory.load_memory_variables({})

输出如下：

{'history': "Human: Hi\nAI: What's up\nHuman: Not much, just hanging\nAI: Cool"}

可以看到把所有的变量读取出来了。一般情况下，ConversationBufferMemory会将所有的信息读出。

2.2 ConversationBufferWindowMemory

from langchain.memory import ConversationBufferWindowMemory

# 设置对话缓冲窗口的轮数为1
memory = ConversationBufferWindowMemory(k = 1)
memory.save_context({"input": "Hi"},
                    {"output": "What's up"})
memory.save_context({"input": "Not much, just hanging"},
                    {"output": "Cool"})
memory.load_memory_variables({})

输出如下：

{'history': 'Human: Not much, just hanging\nAI: Cool'}

可以看到输出的轮次只有一次对话长度。我们尝试用人机对话进行测试。

memory = ConversationBufferWindowMemory(k = 1)
conversation = ConversationChain(
    llm=llm, 
    memory = memory,
    verbose= False
)

conversation.predict(input="Hi, my name is Andrew")

输出如下：

"Hello Andrew! It's nice to meet you. My name is AI. I have access to a vast amount of information, so feel free to ask me anything you'd like to know!"

再进行对话

conversation.predict(input="What is 1+1?")

输出如下：

'1 + 1 equals 2. Is there anything else you would like to know?'

再进行对话

conversation.predict(input="What is my name?")

输出如下：

"I'm sorry, I do not have access to that information. Is there anything else you would like to ask?"

可以看到，设置对话窗口长度为1的时候，LLM的memory中只保留最后一轮的对话。

2.3 ConversationTokenBufferMemory

首先需要安装包，因为ConversationTokenBufferMemory会调用tiktoken包相关内容。

!pip install tiktoken

from langchain.memory import ConversationTokenBufferMemory


# 设置对话过程中的记忆的长度最大为30个token
memory = ConversationTokenBufferMemory(llm=llm, max_token_limit=30)
memory.save_context({"input": "AI is what?!"},
                    {"output": "Amazing!"})
memory.save_context({"input": "Backpropagation is what?"},
                    {"output": "Beautiful!"})
memory.save_context({"input": "Chatbots are what?"}, 
                    {"output": "Charming!"})
memory.load_memory_variables({})

输出如下：

{'history': 'AI: Beautiful!\nHuman: Chatbots are what?\nAI: Charming!'}

2.4 ConversationSummaryBufferMemory

from langchain.memory import ConversationSummaryBufferMemory


schedule = "There is a meeting at 8am with your product team. \
You will need your powerpoint presentation prepared. \
9am-12pm have time to work on your LangChain \
project which will go quickly because Langchain is such a powerful tool. \
At Noon, lunch at the italian resturant with a customer who is driving \
from over an hour away to meet you to understand the latest in AI. \
Be sure to bring your laptop to show the latest LLM demo."

memory = ConversationSummaryBufferMemory(llm=llm, max_token_limit=100)
memory.save_context({"input": "Hello"}, {"output": "What's up"})
memory.save_context({"input": "Not much, just hanging"},
                    {"output": "Cool"})
memory.save_context({"input": "What is on the schedule today?"}, 
                    {"output": f"{schedule}"})
memory.load_memory_variables({})

输出如下：

{'history': "System: The human and AI have a brief exchange of greetings, and the AI informs the human of the day's schedule, including a meeting, work time on a project, and a lunch appointment with a customer interested in AI technology."}

可以看到，由于设置了最大token的长度为100，使用的是SummaryBuffer，因此LLM会对内存中的段落信息进行摘要存储。再继续对话，打开verbose查看chain细节。

conversation = ConversationChain(
    llm = llm, 
    memory = memory,
    verbose = True
)
conversation.predict(input="What would be a good demo to show?")

输出如下：

> Entering new ConversationChain chain...
Prompt after formatting:
The following is a friendly conversation between a human and an AI. The AI is talkative and provides lots of specific details from its context. If the AI does not know the answer to a question, it truthfully says it does not know.

Current conversation:
System: The human and AI have a brief exchange of greetings, and the AI informs the human of the day's schedule, including a meeting, work time on a project, and a lunch appointment with a customer interested in AI technology.
Human: What would be a good demo to show?
AI:

> Finished chain.
"I suggest showcasing our AI's natural language processing capabilities, as well as its ability to analyze and interpret large datasets for actionable insights. We could also demonstrate its real-time decision-making capabilities and its integration with various platforms and applications. Additionally, we could show its ability to automate repetitive tasks and streamline workflow processes."

尝试读取memory中的信息。

memory.load_memory_variables({})

输出如下：

{'history': "System: The human and AI have a brief exchange of greetings, and the AI informs the human of the day's schedule, including a meeting, work time on a project, and a lunch appointment with a customer interested in AI technology.\nHuman: What would be a good demo to show?\nAI: I suggest showcasing our AI's natural language processing capabilities, as well as its ability to analyze and interpret large datasets for actionable insights. We could also demonstrate its real-time decision-making capabilities and its integration with various platforms and applications. Additionally, we could show its ability to automate repetitive tasks and streamline workflow processes."}

3. 思考

在构建应用时，如果token开销比较花钱，可以选择合适的对话memory进行存储关键信息。