Create app_http.py
Browse files- app_http.py +65 -0
app_http.py
ADDED
@@ -0,0 +1,65 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# 服务器端代码
|
2 |
+
|
3 |
+
# 导入Flask库和其他必要的库
|
4 |
+
from flask import Flask, request, jsonify
|
5 |
+
import threading
|
6 |
+
from flask_cors import CORS
|
7 |
+
import os
|
8 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
9 |
+
|
10 |
+
# 设置可见的GPU设备
|
11 |
+
os.environ["CUDA_VISIBLE_DEVICES"] = '0, 1'
|
12 |
+
|
13 |
+
# 创建Flask应用对象
|
14 |
+
app = Flask(__name__)
|
15 |
+
# 允许跨域请求
|
16 |
+
CORS(app)
|
17 |
+
|
18 |
+
# 加载百川大模型的分词器和模型
|
19 |
+
tokenizer = AutoTokenizer.from_pretrained("baichuan-inc/baichuan-7B", trust_remote_code=True)
|
20 |
+
model = AutoModelForCausalLM.from_pretrained("baichuan-inc/baichuan-7B", device_map="auto", trust_remote_code=True)
|
21 |
+
|
22 |
+
# 创建线程锁和计数器
|
23 |
+
lock = threading.Lock()
|
24 |
+
counter = 0
|
25 |
+
MAX_CONCURRENT_REQUESTS = 50 # 最大并发请求数
|
26 |
+
|
27 |
+
# 定义服务接口的路由和方法
|
28 |
+
@app.route('/baichuan/conversation', methods=['POST'])
|
29 |
+
def conversation():
|
30 |
+
global counter # 使用全局变量
|
31 |
+
|
32 |
+
# 请求过载,返回提示信息
|
33 |
+
if counter >= MAX_CONCURRENT_REQUESTS:
|
34 |
+
return jsonify({'message': '请稍等再试'})
|
35 |
+
|
36 |
+
# 获取线程锁
|
37 |
+
with lock:
|
38 |
+
counter += 1 # 增加计数器
|
39 |
+
|
40 |
+
try:
|
41 |
+
# 接收POST请求的数据
|
42 |
+
question = request.json['question']
|
43 |
+
question += '->' # 添加分隔符
|
44 |
+
|
45 |
+
# 对输入进行分词和编码
|
46 |
+
inputs = tokenizer(question, return_tensors='pt')
|
47 |
+
inputs = inputs.to('cuda:0') # 移动到GPU上
|
48 |
+
|
49 |
+
# 调用模型进行生成
|
50 |
+
pred = model.generate(**inputs, max_new_tokens=1024, repetition_penalty=1.1)
|
51 |
+
text = tokenizer.decode(pred.cpu()[0], skip_special_tokens=True) # 对输出进行解码
|
52 |
+
|
53 |
+
# 返回结果
|
54 |
+
response = {'result': text[len(question):]} # 去掉输入部分
|
55 |
+
return jsonify(response)
|
56 |
+
|
57 |
+
finally:
|
58 |
+
# 释放线程锁并减少计数器
|
59 |
+
with lock:
|
60 |
+
counter -= 1
|
61 |
+
|
62 |
+
# 主函数
|
63 |
+
if __name__ == '__main__':
|
64 |
+
print("Flask服务器已启动")
|
65 |
+
app.run(host='0.0.0.0', port=30908) # 设置主机地址和端口号
|