ai/web_demo3.py

147 lines
5.1 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

import json
import streamlit as st
import torch
from transformers import AutoModel, AutoTokenizer
# from api import get_docs
from server.knowledge_base.kb_doc_api import search_docs
# 设置页面标题、图标和布局
st.set_page_config(
page_title="ChatGLM3-6B 演示",
page_icon=":robot:",
layout="wide"
)
# 设置为模型ID或本地文件夹路径
model_path = "THUDM/chatglm3-6b-32k"
def get_docs(query, top_k, score_threshold, knowledge_base_name):
docs = search_docs(query, knowledge_base_name, top_k, score_threshold)
context1 = "\n".join([doc.page_content for doc in docs])
return context1
@st.cache_resource
def get_model():
tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
model = AutoModel.from_pretrained(model_path, trust_remote_code=True).cuda()
# 多显卡支持,使用下面两行代替上面一行,将num_gpus改为你实际的显卡数量
# from utils import load_model_on_gpus
# model = load_model_on_gpus("THUDM/chatglm3-6b", num_gpus=2)
model = model.eval()
return tokenizer, model
# 加载Chatglm3的model和tokenizer
tokenizer, model = get_model()
system_prompt = """你是一个农业领域的AI助手你需要通过已知信息和历史对话来回答人类的问题。
特别注意:已知信息不一定是正确答案,还要根据对话的历史问题和答案进行思考回答,如果找不到答案,请用自己的知识进行回答,不可以乱回答!!!
答案使用中文。\n"""
# 初始化历史记录和past key values
if "history" not in st.session_state:
st.session_state.history = [{'role': 'system', 'content': system_prompt}]
if "past_key_values" not in st.session_state:
st.session_state.past_key_values = None
if "all_prompt_text" not in st.session_state:
st.session_state.all_prompt_text = []
# 设置max_length、top_p和temperature
max_length = st.sidebar.slider("max_length", 0, 32768, 32768, step=1024)
top_p = st.sidebar.slider("top_p", 0.0, 1.0, 0.8, step=0.01)
temperature = st.sidebar.slider("temperature", 0.0, 1.0, 0.6, step=0.01)
# 清理会话历史
buttonClean = st.sidebar.button("清理会话历史", key="clean")
if buttonClean:
st.session_state.history = []
st.session_state.past_key_values = None
if torch.cuda.is_available():
torch.cuda.empty_cache()
st.rerun()
# 渲染聊天历史记录
for i, message in enumerate(st.session_state.history):
if message["role"] == "user":
with st.chat_message(name="user", avatar="user"):
st.markdown(message["content"])
else:
with st.chat_message(name="assistant", avatar="assistant"):
st.markdown(message["content"])
# 输入框和输出框
with st.chat_message(name="user", avatar="user"):
input_placeholder = st.empty()
with st.chat_message(name="assistant", avatar="assistant"):
message_placeholder = st.empty()
def build_prompt(prompt_text, new_query):
knowledge_base_name = 'agriculture'
context = get_docs(new_query, 3, 1, knowledge_base_name)
print("query", new_query)
print("context", context)
prompt = f"\n\n<已知信息>\n{context}\n</已知信息>\n\n"
prompt += f"\n\n<问题>{prompt_text}</问题>\n\n"
return prompt
# 获取用户输入
prompt_text = st.chat_input("请输入您的问题")
# 如果用户输入了内容,则生成回复
if prompt_text:
input_placeholder.markdown(prompt_text)
history = st.session_state.history
with open('a.json', 'w') as f:
json.dump(history, f, ensure_ascii=False)
past_key_values = st.session_state.past_key_values
new_query = ""
assistants = []
# 历史问题
st.session_state.all_prompt_text.append(prompt_text)
# 历史助手
for h in history:
# 助手
if h['role'] == 'assistant':
assistants.append(h['content'])
# 将历史问题和历史助手组合成新问题
for i in range(len(st.session_state.all_prompt_text)):
new_query += st.session_state.all_prompt_text[i]
if i < len(st.session_state.all_prompt_text) - 1:
new_query += ","
else:
new_query += "?"
if i < len(assistants):
new_query += assistants[i]
history = []
for k, v in enumerate(assistants):
a = assistants[k]
q = st.session_state.all_prompt_text[k]
history.append({'role': 'user', 'content': q})
history.append({'role': 'user', 'content': a})
prompt = build_prompt(prompt_text, new_query)
# TODO
# 控制历史信息、滑动窗口的长度!
for response, history, past_key_values in model.stream_chat(
tokenizer,
prompt,
history,
past_key_values=past_key_values,
max_length=max_length,
top_p=top_p,
temperature=temperature,
return_past_key_values=True,
):
message_placeholder.markdown(response)
# 历史assistant
# print(history)
# 更新历史记录和past key values
st.session_state.history = history
st.session_state.past_key_values = past_key_values