You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
76 lines
2.8 KiB
76 lines
2.8 KiB
import os
|
|
os.environ["CUDA_VISIBLE_DEVICES"] = "2"
|
|
import flask
|
|
from transformers import pipeline
|
|
import redis
|
|
import uuid
|
|
import json
|
|
from threading import Thread
|
|
import time
|
|
import requests
|
|
from flask import request
|
|
from vllm import LLM, SamplingParams
|
|
|
|
app = flask.Flask(__name__)
|
|
pool = redis.ConnectionPool(host='localhost', port=63179, max_connections=100, db=5, password="zhicheng123*")
|
|
redis_ = redis.Redis(connection_pool=pool, decode_responses=True)
|
|
|
|
db_key_query = 'query'
|
|
db_key_result = 'result'
|
|
batch_size = 64
|
|
|
|
sampling_params = SamplingParams(temperature=0.95, top_p=0.7,presence_penalty=1.1,stop="</s>", max_tokens=4096)
|
|
models_path = "/home/majiahui/model-llm/openbuddy-mistral-7b-v13.1"
|
|
llm = LLM(model=models_path, tokenizer_mode="slow")
|
|
|
|
def mistral_vllm_models(texts):
|
|
outputs = llm.generate(texts, sampling_params) # 调用模型
|
|
|
|
generated_text_list = [""] * len(texts)
|
|
|
|
# generated_text_list = ["" if len(i[0]) > 5 else i[0] for i in text_list]
|
|
|
|
for i, output in enumerate(outputs):
|
|
index = output.request_id
|
|
generated_text = output.outputs[0].text
|
|
generated_text_list[int(index)] = generated_text
|
|
|
|
return generated_text_list
|
|
|
|
|
|
def classify(): # 调用模型，设置最大batch_size
|
|
while True:
|
|
if redis_.llen(db_key_query) == 0: # 若队列中没有元素就继续获取
|
|
continue
|
|
# else:
|
|
# query = redis_.lpop(db_key_query).decode('UTF-8') # 获取query的text
|
|
# query_ids = json.loads(query)['id']
|
|
# texts = json.loads(query)['texts'] # 拼接若干text 为batch
|
|
for i in range(min(redis_.llen(db_key_query), batch_size)):
|
|
query = redis_.lpop(db_key_query).decode('UTF-8') # 获取query的text
|
|
query_ids.append(json.loads(query)['id'])
|
|
texts.append(json.loads(query)['text']) # 拼接若干text 为batch
|
|
result = mistral_vllm_models(texts) # 调用模型
|
|
print(result)
|
|
redis_.set(query_ids, json.dumps(result)) # 将模型结果送回队列
|
|
|
|
|
|
@app.route("/predict", methods=["POST"])
|
|
def handle_query():
|
|
texts = request.json["texts"] # 获取用户query中的文本 例如"I love you"
|
|
id_ = str(uuid.uuid1()) # 为query生成唯一标识
|
|
d = {'id': id_, 'texts': texts} # 绑定文本和query id
|
|
redis_.rpush(db_key_query, json.dumps(d)) # 加入redis
|
|
while True:
|
|
result = redis_.get(id_) # 获取该query的模型结果
|
|
if result is not None:
|
|
redis_.delete(id_)
|
|
result_text = {'code': "200", 'resilt': json.loads(result.decode('UTF-8'))}
|
|
break
|
|
return flask.jsonify(result_text) # 返回结果
|
|
|
|
|
|
if __name__ == "__main__":
|
|
t = Thread(target=classify)
|
|
t.start()
|
|
app.run(debug=False, host='0.0.0.0', port=14010)
|