Method _generate

opencompass/models/bailing_api_oc.py:160–220

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 gener

(
        self,
        sess,
        input: Union[str, PromptList],
        max_out_len: int,
    )

        return results

    def _generate(
        self,
        sess,
        input: Union[str, PromptList],
        max_out_len: int,
    ) -> 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.
        """
        if isinstance(input, str):
            messages = [{'role': 'user', 'content': input}]
        else:
            messages = []
            for item in input:
                content = item['prompt']
                if not content:
                    continue
                message = {'content': content}
                if item['role'] == 'HUMAN':
                    message['role'] = 'user'
                elif item['role'] == 'BOT':
                    message['role'] = 'assistant'
                elif item['role'] == 'SYSTEM':
                    message['role'] = 'system'
                else:
                    message['role'] = item['role']
                messages.append(message)
        request = {
            'model': self._model,
            'messages': messages,
            'max_tokens': max_out_len,
        }
        request.update(self.generation_kwargs)
        retry_num = 0
        while retry_num < self.retry:
            try:
                response = self._infer_result(request, sess)
            except ConnectionError:
                time.sleep(random.randint(10, 30))
                retry_num += 1  # retry
                continue
            if response.status_code == 200:
                break  # success
            elif response.status_code == 426:
                retry_num += 1  # retry
            elif response.status_code in [302, 429, 500, 504]:
                time.sleep(random.randint(10, 30))
                retry_num += 1  # retry
            else:
                raise ValueError(f'Status code = {response.status_code}')
        else: