Method convert

modelscope/preprocessors/multi_modal.py:735–759

(self, messages: Dict[str, List[Dict]])

        return enc_chunk

    def convert(self, messages: Dict[str, List[Dict]]) -> str:
        texts = []
        image = []
        messages = messages['messages']
        for turn in messages:
            if turn['role'] == 'system':
                role = ''
            elif turn['role'] == 'user':
                role = 'Human: '
            else:
                role = 'AI: '
            if isinstance(turn['content'], str):
                text = f"{role}{turn['content']}"
                texts.append(text)
            else:
                for t in turn['content']:
                    if isinstance(t, str):
                        text = f'{role}{t}'
                    else:
                        text = f'{role}<|image|>'
                        image.append(t['image'])
                    texts.append(text)
        texts = '\n'.join(texts)
        texts += '\nAI: '
        return image, texts

    def __call__(self, messages: Dict[str, Any],
                 **forward_params) -> Dict[str, Any]: