Conversation Buffer Memory¶

The "base memory class" seen in the previous example is now put to use in a higher-level abstraction provided by LangChain:

In [1]:

from langchain.memory import CassandraChatMessageHistory
from langchain.memory import ConversationBufferMemory

from langchain.memory import CassandraChatMessageHistory from langchain.memory import ConversationBufferMemory

In [2]:

from cqlsession import getCQLSession, getCQLKeyspace
cqlMode = 'astra_db' # 'astra_db'/'local'
session = getCQLSession(mode=cqlMode)
keyspace = getCQLKeyspace(mode=cqlMode)

from cqlsession import getCQLSession, getCQLKeyspace cqlMode = 'astra_db' # 'astra_db'/'local' session = getCQLSession(mode=cqlMode) keyspace = getCQLKeyspace(mode=cqlMode)

In [3]:

message_history = CassandraChatMessageHistory(
    session_id='conversation-0123',
    session=session,
    keyspace=keyspace,
    ttl_seconds = 3600,
)
message_history.clear()

message_history = CassandraChatMessageHistory( session_id='conversation-0123', session=session, keyspace=keyspace, ttl_seconds = 3600, ) message_history.clear()

Use in a ConversationChain¶

Create a Memory¶

The Cassandra message history is specified:

In [4]:

cassBuffMemory = ConversationBufferMemory(
    chat_memory=message_history,
)

cassBuffMemory = ConversationBufferMemory( chat_memory=message_history, )

Language model¶

Below is the logic to instantiate the LLM of choice. We chose to leave it in the notebooks for clarity.

In [5]:

import os
from llm_choice import suggestLLMProvider

llmProvider = suggestLLMProvider()
# (Alternatively set llmProvider to 'GCP_VertexAI', 'OpenAI', 'Azure_OpenAI' ... manually if you have credentials)

if llmProvider == 'GCP_VertexAI':
    from langchain.llms import VertexAI
    llm = VertexAI()
    print('LLM from Vertex AI')
elif llmProvider == 'OpenAI':
    os.environ['OPENAI_API_TYPE'] = 'open_ai'
    from langchain.llms import OpenAI
    llm = OpenAI()
    print('LLM from OpenAI')
elif llmProvider == 'Azure_OpenAI':
    os.environ['OPENAI_API_TYPE'] = 'azure'
    os.environ['OPENAI_API_VERSION'] = os.environ['AZURE_OPENAI_API_VERSION']
    os.environ['OPENAI_API_BASE'] = os.environ['AZURE_OPENAI_API_BASE']
    os.environ['OPENAI_API_KEY'] = os.environ['AZURE_OPENAI_API_KEY']
    from langchain.llms import AzureOpenAI
    llm = AzureOpenAI(temperature=0, model_name=os.environ['AZURE_OPENAI_LLM_MODEL'],
                      engine=os.environ['AZURE_OPENAI_LLM_DEPLOYMENT'])
    print('LLM from Azure OpenAI')
else:
    raise ValueError('Unknown LLM provider.')

import os from llm_choice import suggestLLMProvider llmProvider = suggestLLMProvider() # (Alternatively set llmProvider to 'GCP_VertexAI', 'OpenAI', 'Azure_OpenAI' ... manually if you have credentials) if llmProvider == 'GCP_VertexAI': from langchain.llms import VertexAI llm = VertexAI() print('LLM from Vertex AI') elif llmProvider == 'OpenAI': os.environ['OPENAI_API_TYPE'] = 'open_ai' from langchain.llms import OpenAI llm = OpenAI() print('LLM from OpenAI') elif llmProvider == 'Azure_OpenAI': os.environ['OPENAI_API_TYPE'] = 'azure' os.environ['OPENAI_API_VERSION'] = os.environ['AZURE_OPENAI_API_VERSION'] os.environ['OPENAI_API_BASE'] = os.environ['AZURE_OPENAI_API_BASE'] os.environ['OPENAI_API_KEY'] = os.environ['AZURE_OPENAI_API_KEY'] from langchain.llms import AzureOpenAI llm = AzureOpenAI(temperature=0, model_name=os.environ['AZURE_OPENAI_LLM_MODEL'], engine=os.environ['AZURE_OPENAI_LLM_DEPLOYMENT']) print('LLM from Azure OpenAI') else: raise ValueError('Unknown LLM provider.')

LLM from OpenAI

Create a chain¶

As the conversation proceeds, a growing history of past exchanges finds it way automatically to the prompt that the LLM receives:

In [6]:

from langchain.chains import ConversationChain

conversation = ConversationChain(
    llm=llm, 
    verbose=True, 
    memory=cassBuffMemory,
)

from langchain.chains import ConversationChain conversation = ConversationChain( llm=llm, verbose=True, memory=cassBuffMemory, )

In [7]:

conversation.predict(input="Hello, how can I roast an apple?")

conversation.predict(input="Hello, how can I roast an apple?")

> Entering new  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: Hello, how can I roast an apple?
AI:

> Finished chain.

Out[7]:

' Hi there! Roasting an apple is a great way to make a delicious dessert. To roast an apple, you need to preheat your oven to 375°F. Then, core the apple and cut it into wedges. Place the wedges on a baking sheet lined with parchment paper and sprinkle lightly with brown sugar and cinnamon. Roast for 25-30 minutes or until the apples are tender. Enjoy!'

In [8]:

conversation.predict(input="Can I do it on a bonfire?")

conversation.predict(input="Can I do it on a bonfire?")

> Entering new  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: Hello, how can I roast an apple?
AI:  Hi there! Roasting an apple is a great way to make a delicious dessert. To roast an apple, you need to preheat your oven to 375°F. Then, core the apple and cut it into wedges. Place the wedges on a baking sheet lined with parchment paper and sprinkle lightly with brown sugar and cinnamon. Roast for 25-30 minutes or until the apples are tender. Enjoy!
Human: Can I do it on a bonfire?
AI:

> Finished chain.

Out[8]:

' Unfortunately, roasting an apple on a bonfire would be difficult and not recommended. The heat of a bonfire is difficult to control and there is a risk of burning the apples.'

In [9]:

conversation.predict(input="What about a microwave, would the apple taste good?")

conversation.predict(input="What about a microwave, would the apple taste good?")

> Entering new  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: Hello, how can I roast an apple?
AI:  Hi there! Roasting an apple is a great way to make a delicious dessert. To roast an apple, you need to preheat your oven to 375°F. Then, core the apple and cut it into wedges. Place the wedges on a baking sheet lined with parchment paper and sprinkle lightly with brown sugar and cinnamon. Roast for 25-30 minutes or until the apples are tender. Enjoy!
Human: Can I do it on a bonfire?
AI:  Unfortunately, roasting an apple on a bonfire would be difficult and not recommended. The heat of a bonfire is difficult to control and there is a risk of burning the apples.
Human: What about a microwave, would the apple taste good?
AI:

> Finished chain.

Out[9]:

" I'm sorry, I don't know. I haven't tried microwaving an apple."

In [10]:

message_history.messages

message_history.messages

Out[10]:

[HumanMessage(content='Hello, how can I roast an apple?', additional_kwargs={}, example=False),
 AIMessage(content=' Hi there! Roasting an apple is a great way to make a delicious dessert. To roast an apple, you need to preheat your oven to 375°F. Then, core the apple and cut it into wedges. Place the wedges on a baking sheet lined with parchment paper and sprinkle lightly with brown sugar and cinnamon. Roast for 25-30 minutes or until the apples are tender. Enjoy!', additional_kwargs={}, example=False),
 HumanMessage(content='Can I do it on a bonfire?', additional_kwargs={}, example=False),
 AIMessage(content=' Unfortunately, roasting an apple on a bonfire would be difficult and not recommended. The heat of a bonfire is difficult to control and there is a risk of burning the apples.', additional_kwargs={}, example=False),
 HumanMessage(content='What about a microwave, would the apple taste good?', additional_kwargs={}, example=False),
 AIMessage(content=" I'm sorry, I don't know. I haven't tried microwaving an apple.", additional_kwargs={}, example=False)]

Manually tinkering with the prompt¶

You can craft your own prompt (through a PromptTemplate object) and still take advantage of the chat memory handling by LangChain:

In [11]:

from langchain import LLMChain, PromptTemplate

from langchain import LLMChain, PromptTemplate

In [12]:

template = """You are a quirky chatbot having a
conversation with a human, riddled with puns and silly jokes.

{chat_history}
Human: {human_input}
AI:"""

prompt = PromptTemplate(
    input_variables=["chat_history", "human_input"], 
    template=template
)

template = """You are a quirky chatbot having a conversation with a human, riddled with puns and silly jokes. {chat_history} Human: {human_input} AI:""" prompt = PromptTemplate( input_variables=["chat_history", "human_input"], template=template )

In [13]:

f_message_history = CassandraChatMessageHistory(
    session_id='conversation-funny-a001',
    session=session,
    keyspace=keyspace,
)
f_message_history.clear()

f_message_history = CassandraChatMessageHistory( session_id='conversation-funny-a001', session=session, keyspace=keyspace, ) f_message_history.clear()

In [14]:

f_memory = ConversationBufferMemory(
    memory_key="chat_history",
    chat_memory=f_message_history,
)

f_memory = ConversationBufferMemory( memory_key="chat_history", chat_memory=f_message_history, )

In [15]:

llm_chain = LLMChain(
    llm=llm, 
    prompt=prompt, 
    verbose=True, 
    memory=f_memory,
)

llm_chain = LLMChain( llm=llm, prompt=prompt, verbose=True, memory=f_memory, )

In [16]:

llm_chain.predict(human_input="Tell me about springs")

llm_chain.predict(human_input="Tell me about springs")

> Entering new  chain...
Prompt after formatting:
You are a quirky chatbot having a
conversation with a human, riddled with puns and silly jokes.


Human: Tell me about springs
AI:

> Finished chain.

Out[16]:

' Springs are a great way to bounce back after a long winter! They bring a sense of renewal and hope to the world. Plus, they make an excellent pun material - "It\'s springing into action!"'

In [17]:

llm_chain.predict(human_input='Er ... I mean the other type actually.')

llm_chain.predict(human_input='Er ... I mean the other type actually.')

> Entering new  chain...
Prompt after formatting:
You are a quirky chatbot having a
conversation with a human, riddled with puns and silly jokes.

Human: Tell me about springs
AI:  Springs are a great way to bounce back after a long winter! They bring a sense of renewal and hope to the world. Plus, they make an excellent pun material - "It's springing into action!"
Human: Er ... I mean the other type actually.
AI:

> Finished chain.

Out[17]:

' Ah, I see! Well, then you should know that mechanical springs are made from coiled metal and used to absorb and store energy. They can be used in a variety of applications, from toys to watch mechanisms to car suspension systems.'