目录
问题描述:
随着各种LLMs (Large Language Models)的出现,如何调用各种LLMs的api成为了经常会遇见的问题。
问题解决:
下面仅以生成给定sentence的复述句为例,说明如何应用。
data: 存放在.txt文件中,按行存放,每一行是一个句子
⚠️: 只能在本地运行,不可以在服务器上运行,并且,🪜要打开。
代码实现:
python
import openai
import time
import os
import random
from tqdm import tqdm
# openai.api_key = "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"
# openai.api_key = "sk-huUlpXTcDnvNevXgB1399aCa9dB247F599D81737B7CfD927" # 实验室共有
openai.api_key = "YOUR API KEY" # GPT4.0
openai.api_base = 'https://api.ngapi.top/v1'
def get_response(prompt, temperature=0.5, max_tokens=2048):
completion = openai.ChatCompletion.create(
# model="gpt-3.5-turbo",
model="gpt-4",
temperature=0,
top_p=0,
# max_tokens=max_tokens,
messages=[
{"role": "user", "content": f"{prompt}"}
]
)
return completion
python
def generate_prompt(all_sentences): # all_asps是一个list
prompts = []
for cur_sent in all_sentences: # all_compared_pair是一个list:[asp,sub,obj]
# prompt = "Please generate just one synonymous sentence for the given sentence. The current given sentence is : '{}'".format(cur_sent)
prompt = f"Generate a paraphrased version of the following sentence: {cur_sent}"
prompts.append(prompt)
return prompts
python
import re
def process_sentence(sentence):
# 定义要添加空格的特殊字符, 比如's
special_chars = [',', '.', '\'', ''', '"', '"', '(', ')', '[', ']', '{', '}', ':', ';', '?', '!'] # '-', 因为sub,obj中存在很多以'-'为连字符的sub,obj,所以原始句子中,这部分不可以加空格
# 在特殊字符前添加空格
for char in special_chars:
if char == '(': # 特别的,左括号是在后面加空格
sentence = sentence = re.sub(rf'([{char}])', r'\1 ', sentence)
else:
sentence = re.sub(rf'([{char}])', r' \1', sentence)
return sentence
def get_res(all_sentences, wirte_path):
if os.path.exists(wirte_path):
res_file = open(wirte_path)
begin = len(res_file.readlines()) + 1 # 断点续传
res_file.close()
else:
begin = 0
with open(wirte_path, 'a', buffering=1) as res_file:
count = 0
for p in tqdm(generate_prompt(all_sentences), total=len(all_sentences)):
print(p)
count += 1
if count < begin:
continue
try:
res = get_response(p)['choices'][0]['message']['content']
# print("res", res)
processed_res = process_sentence(res) # 处理成以单词,标点符号等为分隔符,空格
# print("processed_res", processed_res)
if "The server is overloaded or not ready yet" in processed_res:
time.sleep(30)
res_file.write("####" + '\n')
else:
res_file.write(processed_res + '\n')
except Exception as e:
if isinstance(e, openai.error.RateLimitError):
time.sleep(30) # 处理拥挤的函数,就只是睡着而已
res = get_response(p)['choices'][0]['message']['content']
processed_res = process_sentence(res) #
res_file.write(processed_res + '\n')
else:
print(e)
exit(-1)
# print("the number of prompt is :", count)
python
from pdb import set_trace as stop
dataset_list = ['14lap','14res', '15res', '16res']
name_list = ['train', 'test', 'dev']
# name_list = ['dev']
# dataset_list = ['14lap']
for dataset in dataset_list:
for name in name_list:
# print("dataste", dataset)
# print("name", name)
sentence_path = f"/Users/qtxu/Workplace/Chatgpt/ATE_Task/{dataset}/{name}_sentence.txt"
write_path = f'/Users/qtxu/Workplace/Chatgpt/ATE_Task/{dataset}/{name}_paraphrase_sentence.txt'
with open(sentence_path,"r") as fr:
lines = fr.readlines()
all_sentences = []
for line in lines:
all_sentences.append(line)
# print("all_sentences", all_sentences)
get_res(all_sentences, write_path)