AI应用开发基础教程_借助LangChain来调用ChatGPT_API

import tiktoken
encoding = tiktoken.encoding_for_model('gpt-3.5-turbo')
text = "This is a test for tiktoken."
tokens = encoding.encode(text)
print(len(text)) # 28
print(tokens) # [2028, 374, 264, 1296, 369, 87272, 5963, 13]
print(len(tokens)) # 8

import streamlit as st
from langchain.chat_models import ChatOpenAI
from langchain.schema import (SystemMessage, HumanMessage, AIMessage)


def main():
    llm = ChatOpenAI(temperature=0)

    st.set_page_config(
        page_title="My Great ChatGPT",
        page_icon="🤗"
    )
    st.header("My Great ChatGPT 🤗")

    # Initialize chat history
    if "messages" not in st.session_state:
        st.session_state.messages = [
            SystemMessage(content="You are a helpful assistant.")
        ]

    # Monitor user input
    if user_input := st.chat_input("Input your question here:"):
        st.session_state.messages.append(HumanMessage(content=user_input))
        with st.spinner("ChatGPT is typing ..."):
            response = llm(st.session_state.messages)
        st.session_state.messages.append(AIMessage(content=response.content))

    # Display chat history
    messages = st.session_state.get('messages', [])
    for message in messages:
        if isinstance(message, AIMessage):
            with st.chat_message('assistant'):
                st.markdown(message.content)
        elif isinstance(message, HumanMessage):
            with st.chat_message('user'):
                st.markdown(message.content)
        else:  # isinstance(message, SystemMessage):
            st.write(f"System message: {message.content}")


if __name__ == '__main__':
    main()

import streamlit as st
from langchain.chat_models import ChatOpenAI
from langchain.schema import (SystemMessage, HumanMessage, AIMessage)


def main():
    llm = ChatOpenAI(temperature=0)

    st.set_page_config(
        page_title="我的伟大的ChatGPT",
        page_icon="🤗"
    )
    st.header("我的伟大的ChatGPT 🤗")

    # 初始化聊天历史
    if "messages" not in st.session_state:
        st.session_state.messages = [
            SystemMessage(content="你是一个有用的助手。")
        ]

    container = st.container()
    with container:
        with st.form(key='my_form', clear_on_submit=True):
            user_input = st.text_area(label='Message: ', key='input', height=100)
            submit_button = st.form_submit_button(label='Send')
    if submit_button and user_input:
        st.session_state.messages.append(HumanMessage(content=user_input))
        with st.spinner("ChatGPT正在输入..."):
            response = llm(st.session_state.messages)
        st.session_state.messages.append(AIMessage(content=response.content))

    # 显示聊天历史
    messages = st.session_state.get('messages', [])
    for message in messages:
        if isinstance(message, AIMessage):
            with st.chat_message('assistant'):
                st.markdown(message.content)
        elif isinstance(message, HumanMessage):
            with st.chat_message('user'):
                st.markdown(message.content)
        else:  # isinstance(message, SystemMessage):
            st.write(f"系统消息：{message.content}")


if __name__ == '__main__':
    main()

from langchain.chat_models import ChatOpenAI
from langchain.schema import (
  SystemMessage, # System message
  HumanMessage, # Human question
  AIMessage # ChatGPT's response
)
llm = ChatOpenAI() # Feature that calls the ChatGPT API
message = "Hi, ChatGPT!" # Write your question here
messages = [
  SystemMessage(content="You are a helpful assistant."),
  HumanMessage(content=message)
]
response = llm(messages)
print(response)

message = "Hi, ChatGPT!" # Write your question here
messages = [
	SystemMessage(content="Please Reply in Japanese"),
	HumanMessage(content=message)
]
response = llm(messages)
print(response)
# content='こんにちは、何かお⼿伝いできますか？' additional_kwargs={} example=False

message = "Writing a book about creating an AI app with ChatGPT and Streamlit. Let's come up with one
title."
messages = [
	SystemMessage(content="You are a helpful assistant."),
	HumanMessage(content=message)
]
for temperature in [0, 1, 2]:
  print(f'==== temp: {temperature}')
  llm = ChatOpenAI(temperature=temperature)
  for i in range(3):
	  print(llm(messages).content)

# Initialize chat history
if "messages" not in st.session_state:
  st.session_state.messages = [
    SystemMessage(content="You are a helpful assistant.")
  ]
# ...
# Monitor user input
if user_input := st.chat_input("Input your question here:"):
  st.session_state.messages.append(HumanMessage(content=user_input))
  with st.spinner("ChatGPT is typing ..."):
    response = llm(st.session_state.messages)
  st.session_state.messages.append(AIMessage(content=response.content))
# Display chat history
messages = st.session_state.get('messages', [])
for message in messages:
  ... # Code for displaying chat history

messages = st.session_state.get('messages', [])
for message in messages:
    if isinstance(message, AIMessage):
        with st.chat_message('assistant'):
            st.markdown(message.content)
    elif isinstance(message, HumanMessage):
        with st.chat_message('user'):
            st.markdown(message.content)
    else: # isinstance(message, SystemMessage):
        st.write(f"系统消息：{message.content}")

import streamlit as st
from langchain.chat_models import ChatOpenAI
from langchain.schema import (
    SystemMessage,
    HumanMessage,
    AIMessage
)
from langchain.callbacks import get_openai_callback


def init_page():
    st.set_page_config(
        page_title="My Great ChatGPT",
        page_icon="🤗"
    )
    st.header("My Great ChatGPT 🤗")
    st.sidebar.title("Options")


def init_messages():
    clear_button = st.sidebar.button("Clear Conversation", key="clear")
    if clear_button or "messages" not in st.session_state:
        st.session_state.messages = [
            SystemMessage(content="You are a helpful assistant.")
        ]
        st.session_state.costs = []


def select_model():
    model = st.sidebar.radio("Choose a model:", ("GPT-3.5", "GPT-4"))
    if model == "GPT-3.5":
        model_name = "gpt-3.5-turbo"
    else:
        model_name = "gpt-4"

    # Add a slider to allow users to select the temperature from 0 to 2.
    # The initial value should be 0.0, with an increment of 0.01.
    temperature = st.sidebar.slider("Temperature:", min_value=0.0, max_value=2.0, value=0.0, step=0.01)

    return ChatOpenAI(temperature=temperature, model_name=model_name)


def get_answer(llm, messages):
    with get_openai_callback() as cb:
        answer = llm(messages)
    return answer.content, cb.total_cost


def main():
    init_page()

    llm = select_model()
    init_messages()

    # Monitor user input
    if user_input := st.chat_input("Input your question here:"):
        st.session_state.messages.append(HumanMessage(content=user_input))
        with st.spinner("ChatGPT is typing ..."):
            answer, cost = get_answer(llm, st.session_state.messages)
        st.session_state.messages.append(AIMessage(content=answer))
        st.session_state.costs.append(cost)

    messages = st.session_state.get('messages', [])
    for message in messages:
        if isinstance(message, AIMessage):
            with st.chat_message('assistant'):
                st.markdown(message.content)
        elif isinstance(message, HumanMessage):
            with st.chat_message('user'):
                st.markdown(message.content)
        else:  # isinstance(message, SystemMessage):
            st.write(f"System message: {message.content}")

    costs = st.session_state.get('costs', [])
    st.sidebar.markdown("## Costs")
    st.sidebar.markdown(f"**Total cost: ${sum(costs):.5f}**")
    for cost in costs:
        st.sidebar.markdown(f"- ${cost:.5f}")

if __name__ == '__main__':
    main()

# 导入相关类库
import streamlit as st

# 在侧边栏显示标题
st.sidebar.title("Options")
# 在侧边栏添加选项按钮
model = st.sidebar.radio("Choose a model:", ("GPT-3.5", "GPT-4"))
# 在侧边栏添加一个按钮
clear_button = st.sidebar.button("Clear Conversation", key="clear")
# 在侧边栏添加一个滑块以从0到2选择温度
# 初始值设为0.0，增量为0.1
temperature = st.sidebar.slider("Temperature:", min_value=0.0, max_value=2.0, value=0.0, step=0.1)
# Streamlit将以HTML形式显示markdown内容
# （你也可以在主窗口中使用它）
st.sidebar.markdown("## Costs")
st.sidebar.markdown("**Total cost**")
for i in range(3):
    st.sidebar.markdown(f"- ${i+0.01}") # dummy

def select_model():
    model = st.sidebar.radio("Choose a model:", ("GPT-3.5", "GPT-4"))
    if model == "GPT-3.5":
        model_name = "gpt-3.5-turbo"
    else:
        model_name = "gpt-4"
    # Add a slider to allow users to select the temperature from 0 to 2.
    # The initial value should be 0.0, with an increment of 0.01.
    temperature = st.sidebar.slider("Temperature:", min_value=0.0, max_value=2.0, value=0.0, step=0.01)
    return ChatOpenAI(temperature=temperature, model_name=model_name)

def init_messages():
    clear_button = st.sidebar.button("清除对话", key="clear")
    if clear_button or "messages" not in st.session_state:
        st.session_state.messages = [SystemMessage(content="您是一个有用的助手。")]
    st.session_state.costs = []
def main():
    init_page()
    Ilm = select_model()
    init_messages()

from langchain.llms import OpenAI
from langchain.callbacks import get_openai_callback
llm = OpenAI(model_name="text-davinci-002")
with get_openai_callback() as cb:
    result = llm("Tell me a joke")
    print(cb)
    print(cb.total_cost)
    print(cb.total_tokens)
    print(cb.prompt_tokens)
    print(cb.completion_tokens)

Tokens Used: 23
        Prompt Tokens: 4
        Completion Tokens: 19
Successful Requests: 1
Total Cost (USD): $0.00046
0.00046
23
4
19

# The past history will be displayed normally.
messages = st.session_state.get('messages', [])
for message in messages:
    if isinstance(message, AIMessage):
        with st.chat_message('assistant'):
            st.markdown(message.content)
    elif isinstance(message, HumanMessage):
        with st.chat_message('user'):
            st.markdown(message.content)
    else: # isinstance(message, SystemMessage):
        st.write(f"System message: {message.content}")

# Display the latest exchange separately (a bit more complex).
if user_input := st.chat_input("Input your question here:"):
    st.session_state.messages.append(
        HumanMessage(content=user_input))
    st.chat_message("user").markdown(user_input)
    with st.chat_message("assistant"):
        st_callback = StreamlitCallbackHandler(
            st.container())
        response = llm(messages, callbacks=[st_callback])
    st.session_state.messages.append(
        AIMessage(content=response.content))

[theme] 
base="dark"

def get_content(url):
    try:
        with st.spinner("Fetching Content ..."):
            response = requests.get(url)
            soup = BeautifulSoup(response.text, 'html.parser')
            # fetch text from main (change the below code to filter page)
            if soup.main:
                return soup.main.get_text()
            elif soup.article:
                return soup.article.get_text()
            else:
                return soup.body.get_text()
    except:
        st.write('something wrong')
        return None

def build_prompt(content, n_chars=300):
    return f"""Here is the content of a web page. Please provide a concise summary of around {n_chars} characters.

========

{content[:1000]}

"""


def get_answer(llm, messages):
    with get_openai_callback() as cb:
        answer = llm(messages)
    return answer.content, cb.total_cost

with response_container:
  st.markdown("## Summary")
  st.write(answer)
  st.markdown("---")
  st.markdown("## Original Text")
  st.write(content)

prompt_template = f"""
Using the following background knowledge, please answer the user's questions.
=====
Background knowledge
・{Related knowledge fetched from DB-1}
・{Related knowledge fetched from DB-2}
・{Related knowledge fetched from DB-3}
=====
User's question
・{Question from the user}
"""
# Example of the prompt
prompt_example = f"""
Using the following background knowledge, please answer the user's questions.
=====
Background knowledge
・As of June 23, 2023, Shohei Ohtani has hit 24 home runs
・As of June 23, 2023, Shohei Ohtani has achieved 6 wins
・Shohei Ohtani's height is 193cm
=====
User's question
・Please tell me about Shohei Ohtani's player performance in 2023
"""
# >> "As a batter, Shohei Ohtani has hit 24 home runs, and as a pitcher, he has achieved 6 wins."

from PyPDF2 import PdfReader
def get_pdf_text():
    uploaded_file = st.file_uploader(
        label='Upload your PDF here😇',
        type='pdf'
    )
    if uploaded_file:
        pdf_reader = PdfReader(uploaded_file)
        text = '\n\n'.join([page.extract_text() for page in pdf_reader.pages])
        text_splitter = RecursiveCharacterTextSplitter.from_tiktoken_encoder(
            model_name=st.session_state.emb_model_name,
            chunk_size=250,
            chunk_overlap=0,
        )
        return text_splitter.split_text(text)
    else:
        return None

from langchain.embeddings import OpenAIEmbeddings
emb_model = OpenAIEmbeddings()
text = "Hello World!"
result = emb_model.embed_documents([text])
print(f'Embedding dimentions: {len(result[0])}')
print(result[0][:5])

from qdrant_client import QdrantClient 
from langchain.vectorstores import Qdrant 
from langchain.embeddings.openai import OpenAIEmbeddings 
QDRANT_PATH = "./local_qdrant" 
COLLECTION_NAME = "my_collection"

def load_qdrant():
    client = QdrantClient(path=QDRANT_PATH)
    # Get all collection names.
    collections = client.get_collections().collections
    collection_names = [collection.name for collection in collections]
    # If the collection does not exist, create it.
    if COLLECTION_NAME not in collection_names:
        client.create_collection(
            collection_name=COLLECTION_NAME,
            vectors_config=VectorParams(size=1536, distance=Distance.COSINE),
        )
        print('collection created')
    return Qdrant(
        client=client,
        collection_name=COLLECTION_NAME,
        embeddings=OpenAIEmbeddings()
    )
def build_vector_store(pdf_text):
    qdrant = load_qdrant()
    qdrant.add_texts(pdf_text)

def page_pdf_upload_and_build_vector_db():
    st.title("PDF Upload")
    container = st.container()
    with container:
        # The one we made earlier.
        pdf_text = get_pdf_text()
        if pdf_text:
            with st.spinner("Loading PDF ..."):
                build_vector_store(pdf_text)

page_pdf_upload_and_build_vector_db()

from qdrant_client import QdrantClient
# Saving to local storage
client = QdrantClient(path="./local_qdrant")
# Saving to qdrant cloud (we'll discuss it in detail in the next chapter).
client = QdrantClient(
    url="https://my-qdrant-db.us-east-1-0.aws.cloud.qdrant.io:6333",
    api_key="api-key-hoge123fuga456"
)

# Get all collection names.
collections = client.get_collections().collections
collection_names = [collection.name for collection in collections]
# If the collection does not exist, create it.
if COLLECTION_NAME not in collection_names:
    client.create_collection(
        collection_name=COLLECTION_NAME,
        vectors_config=VectorParams(size=1536, distance=Distance.COSINE),
    )
    print('collection created')

def load_qdrant():
    return Qdrant(
        client=client,
        collection_name=COLLECTION_NAME,
        embeddings=OpenAlEmbeddings()
    )
def build_vector_store(pdf_text):
    qdrant = load_qdrant()

def build_vector_store(pdf_text): 
  qdrant = load_qdrant() 
  qdrant.add_texts(pdf_text) 

Qdrant.from_texts(
  pdf_text,
  OpenAIEmbeddings(),
  path="./local_qdrant",
  collection_name="my_documents",
)

% sqlite3 collection/my_documents/storage.sqlite
SQLite version 3.38.2 2022-03-26 13:51:10
Enter ".help" for usage hints.
sqlite> .tables
points
sqlite> .schema points
CREATE TABLE points (id TEXT PRIMARY KEY, point BLOB);
sqlite> select * from points;
gASVJAAAAAAAAACMIDNkYmExYjYwZTJiYjQzMGM5NmI0YzJjNWJlMTBlMDU4lC4=|V8
gASVJAAAAAAAAACMIDY4NjdmMzY5MDYxNjQ2NzJhNGRmNTVkNDY1NThjNjNmlC4=|9
gASVJAAAAAAAAACMIDY5OWYzZWRhODc3YjRlZTJhYTNlZDQ2YzVjODUxMzQ4lC4=|8
sqlite> select count(*) from points;
235

import qdrant_client
from langchain.vectorstores import Qdrant
from langchain.embeddings.openai import OpenAIEmbeddings
client = qdrant_client.QdrantClient(path="./local_qdrant")
qdrant = Qdrant(
	client=client,
	collection_name="my_collection",
	embeddings=OpenAIEmbeddings()
)
query = "Regarding Toyota's financial results."
docs = qdrant.similarity_search(query=query, k=2)
for i in docs:
	print({"content": i.page_content, "metadata": i.metadata})

from qdrant_client import QdrantClient
from langchain.vectorstores import Qdrant
from langchain.embeddings.openai import OpenAIEmbeddings
QDRANT_PATH = "./local_qdrant"
COLLECTION_NAME = "my_collection"
def load_qdrant():
    """
    它与上一章创建的内容完全相同，所以我将省略它.
    client = QdrantClient()
    ...
def build_qa_model(Ilm):
    qdrant = load_qdrant()
    ...
    )
def ask(qa, query):
    with get_openai_callback() as cb:
        ...

def build_qa_model(llm):
    qdrant = load_qdrant()
    retriever = qdrant.as_retriever(
        # There are also "mmr," "similarity_score_threshold," and others.
        search_type="similarity",
        # How many documents to retrieve? (default: 4)
        search_kwargs={"k":10}
    )
    return RetrievalQA.from_chain_type(
        llm=llm,
        chain_type="stuff", 
        retriever=retriever,
        return_source_documents=True,
        verbose=True
    )

query = "Let me briefly tell you about Toyota's latest financial results."
qa = build_qa_model(llm)
answer = qa.run(query)

# Here's an example of using the similarity_score_threshold.
retriever = db.as_retriever(
  search_type="similarity_score_threshold",
  search_kwargs={"score_threshold": .5}
)
# Here's an example of limiting the context of retrieval using filters.
retriever = db.as_retriever(
  search_kwargs={
  	"filter": {"company": "toyota"}
  }
)

#
https://github.com/hwchase17/langchain/blob/master/langchain/chains/question_answering/stuff_prompt.py
prompt_template = """Use the following pieces of context to answer the question at the end.
If you don't know the answer, just say that you don't know,
don't try to make up an answer.
{context}
Question: {question}
Helpful Answer:"""

prompt_template = """Use the following pieces of context ... Answer in Italian:"""
PROMPT = PromptTemplate(
	template=prompt_template, input_variables=["context", "question"]
)
qa = RetrievalQA.from_chain_type(
  llm=llm
  chain_type="stuff",
  retriever=qdrant.as_retriever(),
  # Specify here
  chain_type_kwargs={"prompt": PROMPT}
)
query = "What did the president say about Ketanji Brown Jackson"
qa.run(query)
" Il presidente ha detto che Ketanji Brown Jackson è una delle menti legali più importanti del paese, che
continuerà l'eccellenza di Justice Breyer e che ha ricevuto un ampio sostegno, da Fraternal Order of Police
a ex giudici nominati da democratici e repubblicani."

curl \
    -X GET 'https://xxxxxxx-xxxx-xxxx-xxxx-434eab336aa0.us-east4-0.gcp.cloud.qdrant.io:6333' \
    --header 'api-key: xxxxxxxxxxxxxxxxxxxxxxxxx'

def load_qdrant():
  # This is what I had written before.: client = QdrantClient(path=QDRANT_PATH)
  # Get url, api_key from Qdrant Cloud dashboard
  client = QdrantClient(
    url="https://hogehogehogehoge.us-east-1-0.aws.cloud.qdrant.io:6333",
    api_key="api-key-hogehogehogehoge"
  )

def load_qdrant():
  client = QdrantClient(
 	 url=os.environment['QDRANT_CLOUD_ENDPOINT'],
	 api_key=os.environment['QDRANT_CLOUD_API_KEY']
  )