Class OpenAI

opencompass/models/openai_api.py:32–559

Model wrapper around OpenAI's models. Args: path (str): The name of OpenAI's model. 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.

@MODELS.register_module()
class OpenAI(BaseAPIModel):
    """Model wrapper around OpenAI's models.

    Args:
        path (str): The name of OpenAI's model.
        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.
        query_per_second (int): The maximum queries allowed per second
            between two consecutive calls of the API. Defaults to 1.
        retry (int): Number of retires if the API call fails. Defaults to 2.
        key (str or List[str]): OpenAI key(s). In particular, when it
            is set to "ENV", the key will be fetched from the environment
            variable $OPENAI_API_KEY, as how openai defaults to be. If it's a
            list, the keys will be used in round-robin manner. Defaults to
            'ENV'.
        org (str or List[str], optional): OpenAI organization(s). If not
            specified, OpenAI uses the default organization bound to each API
            key. If specified, the orgs will be posted with each request in
            round-robin manner. Defaults to None.
        meta_template (Dict, optional): The model's meta prompt
            template if needed, in case the requirement of injecting or
            wrapping of any meta instructions.
        openai_api_base (str): The base url of OpenAI's API. Defaults to
            'https://api.openai.com/v1/chat/completions'.
        openai_proxy_url (str, optional): An optional proxy url to use when
            connecting to OpenAI's API. When set to 'ENV', the url will be
            fetched from the environment variable $OPENAI_PROXY_URL.
            Defaults to None.
        mode (str, optional): The method of input truncation when input length
            exceeds max_seq_len. 'front','mid' and 'rear' represents the part
            of input to truncate. Defaults to 'none'.
        temperature (float, optional): What sampling temperature to use.
            If not None, will override the temperature in the `generate()`
            call. Defaults to None.
        tokenizer_path (str, optional): The path to the tokenizer. Use path if
            'tokenizer_path' is None, otherwise use the 'tokenizer_path'.
            Defaults to None.
        extra_body (Dict, optional): Add additional JSON properties to
            the request
        think_tag (str, optional): The tag to use for reasoning content.
            Defaults to '</think>'.
    """

    is_api: bool = True

    def __init__(
        self,
        path: str = 'gpt-3.5-turbo',
        max_seq_len: int = 16384,
        query_per_second: int = 1,
        rpm_verbose: bool = False,
        retry: int = 2,
        key: Union[str, List[str]] = 'ENV',
        org: Optional[Union[str, List[str]]] = None,
        meta_template: Optional[Dict] = None,
        openai_api_base: str = OPENAI_API_BASE,
        openai_proxy_url: Optional[str] = None,