星尘

使用 Streamlit 搭建简单 GPT 页面

发表于 更新于 分类于 ，

介绍

Streamlit 是一个开源的 Python 框架，用于快速构建和共享用于机器学习和数据科学项目的交互式 Web 应用程序。它简化了应用程序的开发过程，允许开发者通过编写简单的 Python 脚本来创建功能丰富的数据可视化界面 ——by deepseek

示例

运行国内训练过的 LLama3 中文模型

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
import streamlit as st
from langchain_core.prompts import ChatPromptTemplate
from langchain_community.chat_models import ChatOpenAI
from langchain_community.chat_models import ChatOllama

st.title("极简对话")

prompt = ChatPromptTemplate.from_messages([
    ("system", "请用专业的术语解释下面问题，不超过100个字"),
    ("human", "{topic}"),
])

# model = ChatOpenAI(
#     model='deepseek-chat',
#     openai_api_key='sk-442c0bee742143a5bca92ae86a6b2ec6',
#     openai_api_base='https://api.deepseek.com',
#     max_tokens=1024
# )

model = ChatOllama(model="llamafamily/llama3-chinese-8b-instruct")
chain = prompt | model

with st.form("form"):
    text = st.text_area("请输入问题: ")
    submit = st.form_submit_button("提交")
    if submit:
        st.info(chain.invoke({"topic": text}))
    chain.get_graph().print_ascii()

效果

Chain 调用情况

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
    +-------------+    
    | PromptInput |    
    +-------------+    
           *           
           *           
           *           
+--------------------+ 
| ChatPromptTemplate | 
+--------------------+ 
           *           
           *           
           *           
    +------------+     
    | ChatOpenAI |     
    +------------+     
           *           
           *           
           *           
 +------------------+  
 | ChatOpenAIOutput |  
 +------------------+

加载 tools 实现在线查询

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
import streamlit as st
from langchain.agents import initialize_agent, AgentType, load_tools
from langchain_openai import OpenAI, ChatOpenAI

open_api_key = st.sidebar.text_input("OpenAI API Key")

if prompt := st.chat_input():
    if not open_api_key:
        st.info("请输入 OpenAI Key")
        st.stop()
    model = ChatOpenAI(
        model='deepseek-chat',
        openai_api_key=open_api_key,
        openai_api_base='https://api.deepseek.com',
        max_tokens=1024,
        streaming=True
    )
    tools = load_tools(["ddg-search"])
    agent = initialize_agent(tools, model, agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION, verbose=True)
    st.chat_message("user").write(prompt)

    with st.chat_message("assistant"):
        response = agent.run(prompt)
        st.write(response)

langchain 执行信息如下