File size: 4,810 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 |
import threading
from concurrent.futures import ThreadPoolExecutor
from typing import Dict, List, Optional, Union
from opencompass.models.base import BaseModel, LMTemplateParser
from opencompass.utils.logging import get_logger
from opencompass.utils.prompt import PromptList
PromptType = Union[PromptList, str]
def valid_str(string, coding='utf-8'):
"""decode text according to its encoding type."""
invalid_chars = [b'\xef\xbf\xbd']
bstr = bytes(string, coding)
for invalid_char in invalid_chars:
bstr = bstr.replace(invalid_char, b'')
ret = bstr.decode(encoding=coding, errors='ignore')
return ret
class TurboMindAPIModel(BaseModel):
"""Model wrapper for TurboMind Triton Inference Server gRPC API.
Args:
path (str): The name of OpenAI's model.
tis_addr (str): The address (ip:port format) of turbomind's
triton inference server
max_seq_len (int): The maximum allowed sequence length of a model.
Note that the length of prompt + generated tokens shall not exceed
this value. Defaults to 2048.
meta_template (Dict, optional): The model's meta prompt
template if needed, in case the requirement of injecting or
wrapping of any meta instructions.
"""
is_api: bool = True
def __init__(
self,
path: str,
api_addr: str = 'http://0.0.0.0:23333',
max_seq_len: int = 2048,
meta_template: Optional[Dict] = None,
):
super().__init__(path=path,
max_seq_len=max_seq_len,
meta_template=meta_template)
from lmdeploy.serve.openai.api_client import APIClient
self.chatbot = APIClient(api_addr)
self.model_name = self.chatbot.available_models[0]
self.logger = get_logger()
self.template_parser = LMTemplateParser(meta_template)
self.eos_token_id = None
if meta_template and 'eos_token_id' in meta_template:
self.eos_token_id = meta_template['eos_token_id']
self.api_addr = api_addr
def generate(
self,
inputs: List[str or PromptList],
max_out_len: int = 512,
temperature: float = 1.0,
) -> 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.
temperature (float): What sampling temperature to use,
between 0 and 2. Higher values like 0.8 will make the output
more random, while lower values like 0.2 will make it more
focused and deterministic. Defaults to 0.7.
Returns:
List[str]: A list of generated strings.
"""
with ThreadPoolExecutor() as executor:
results = list(
executor.map(self._generate, inputs,
[max_out_len] * len(inputs),
[temperature] * len(inputs)))
return results
def get_token_len(self, prompt: str) -> int:
input_ids, length = self.chatbot.encode(prompt)
return length
def wait(self):
"""Wait till the next query can be sent.
Applicable in both single-thread and multi-thread environments.
"""
return self.token_bucket.get_token()
def _generate(self, prompt: str or PromptList, max_out_len: int,
temperature: float) -> str:
"""Generate results given a list of inputs.
Args:
prompt (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.
temperature (float): What sampling temperature to use,
between 0 and 2. Higher values like 0.8 will make the output
more random, while lower values like 0.2 will make it more
focused and deterministic.
Returns:
str: The generated string.
"""
assert type(
prompt) is str, 'We only support string for TurboMind RPC API'
response = ''
for output in self.chatbot.completions_v1(
session_id=threading.currentThread().ident,
prompt=prompt,
model=self.model_name,
max_tokens=max_out_len,
temperature=temperature,
top_p=0.8,
top_k=1):
response += output['choices'][0]['text']
response = valid_str(response)
return response
|