File size: 5,783 Bytes
256a159 |
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 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 |
import time
from concurrent.futures import ThreadPoolExecutor
from typing import Dict, List, Optional, Union
from opencompass.utils.prompt import PromptList
from .base_api import BaseAPIModel
PromptType = Union[PromptList, str]
class Qwen(BaseAPIModel):
"""Model wrapper around Qwen.
Documentation:
https://help.aliyun.com/zh/dashscope/developer-reference/tongyi-thousand-questions/
Args:
path (str): The name of qwen model.
e.g. `qwen-max`
key (str): Authorization key.
query_per_second (int): The maximum queries allowed per second
between two consecutive calls of the API. Defaults to 1.
max_seq_len (int): Unused here.
meta_template (Dict, optional): The model's meta prompt
template if needed, in case the requirement of injecting or
wrapping of any meta instructions.
retry (int): Number of retires if the API call fails. Defaults to 2.
"""
def __init__(self,
path: str,
key: str,
query_per_second: int = 1,
max_seq_len: int = 2048,
meta_template: Optional[Dict] = None,
retry: int = 5,
generation_kwargs: Dict = {}):
super().__init__(path=path,
max_seq_len=max_seq_len,
query_per_second=query_per_second,
meta_template=meta_template,
retry=retry,
generation_kwargs=generation_kwargs)
import dashscope
dashscope.api_key = key
self.dashscope = dashscope
def generate(
self,
inputs: List[str or PromptList],
max_out_len: int = 512,
) -> List[str]:
"""Generate results given a list of inputs.
Args:
inputs (List[str or PromptList]): A list of strings or PromptDicts.
The PromptDict should be organized in OpenCompass'
API format.
max_out_len (int): The maximum length of the output.
Returns:
List[str]: A list of generated strings.
"""
with ThreadPoolExecutor() as executor:
results = list(
executor.map(self._generate, inputs,
[max_out_len] * len(inputs)))
self.flush()
return results
def _generate(
self,
input: str or PromptList,
max_out_len: int = 512,
) -> str:
"""Generate results given an input.
Args:
inputs (str or PromptList): A string or PromptDict.
The PromptDict should be organized in OpenCompass'
API format.
max_out_len (int): The maximum length of the output.
Returns:
str: The generated string.
"""
assert isinstance(input, (str, PromptList))
"""
{
"messages": [
{"role":"user","content":"请介绍一下你自己"},
{"role":"assistant","content":"我是通义千问"},
{"role":"user","content": "我在上海，周末可以去哪里玩？"},
{"role":"assistant","content": "上海是一个充满活力和文化氛围的城市"},
{"role":"user","content": "周末这里的天气怎么样？"}
]
}
"""
if isinstance(input, str):
messages = [{'role': 'user', 'content': input}]
else:
messages = []
for item in input:
msg = {'content': item['prompt']}
if item['role'] == 'HUMAN':
msg['role'] = 'user'
elif item['role'] == 'BOT':
msg['role'] = 'assistant'
elif item['role'] == 'SYSTEM':
msg['role'] = 'system'
messages.append(msg)
data = {'messages': messages}
data.update(self.generation_kwargs)
max_num_retries = 0
while max_num_retries < self.retry:
self.acquire()
try:
response = self.dashscope.Generation.call(
model=self.path,
**data,
)
except Exception as err:
print('Request Error:{}'.format(err))
time.sleep(1)
continue
self.release()
if response is None:
print('Connection error, reconnect.')
# if connect error, frequent requests will casuse
# continuous unstable network, therefore wait here
# to slow down the request
self.wait()
continue
if response.status_code == 200:
try:
msg = response.output.text
return msg
except KeyError:
print(response)
self.logger.error(str(response.status_code))
time.sleep(1)
continue
if response.status_code == 429:
print('Rate limited')
time.sleep(2)
continue
if response.status_code == 400:
msg = 'Output data may contain inappropriate content.'
return msg
if ('Range of input length should be ' in response.message
or # input too long
'Input data may contain inappropriate content.'
in response.message): # bad input
print(response.message)
return ''
print(response)
max_num_retries += 1
raise RuntimeError(response.message)
|