def gen_struct_qa()

in src/models/struxgpt_v2.py [0:0]

• 42 lines of code
• 10 McCabe index (conditional complexity)

    def gen_struct_qa(self, qa_gen_template: str,
                            ctx_struct: Union[StructItem, dict], 
                            chunk_titles: List[str], chunk_contents: List[str], idx: str, 
                            demonstrations: str = "", response: str = "",
                            prompt_only: bool = False, parse_only: bool = False,
                            **prompt_kwargs):
        if not parse_only:
            #### generate qa data
            ctx_struct = self.check_struct_item(ctx_struct, StructItem)
            ctx_mindmap = self.dump_mindmap(ctx_struct, highlights=chunk_titles)
            contents = ''
            for title, content in zip(chunk_titles, chunk_contents):
                contents += f'\n* **{title}**\n```\n{content}\n```\n'
            chunk_num = len(chunk_titles)

            prompt = qa_gen_template.format(mindmap=ctx_mindmap, field=ctx_struct.scope, 
                                            passage_num=chunk_num, contents=contents,
                                            demos=demonstrations, **prompt_kwargs)
            if prompt_only:
                return {'idx': idx, 'question': prompt, 
                        'ctx_struct': ctx_struct, 'ctx_mindmap': ctx_mindmap}
            else:
                response = self.call_model(prompt)
        
        #### parse qa data
        try:
            invalid_keywords = ['test', 'exam', 'mindmap', 'segment',
                                '考', '思维导图', '片段']
            sections = response.strip().split('\n\n')
            assert len(sections) == 3
            assert sections[0].splitlines()[0].startswith('# Question')
            assert sections[1].splitlines()[0].startswith('# Answer')
            assert sections[2].splitlines()[0].startswith('# Explanation')
            question, answer, explanation = [
                '\n'.join(section.splitlines()[1:]) for section in sections
            ]
            explanation = '\n'.join(line for line in explanation.splitlines() \
                                        if all(flag not in line.lower() for flag in invalid_keywords))
            for flag in ['**Integrating Mindmap Elements:**', '**Logical Deduction:**', '**Conclusion:**']:
                explanation = explanation.replace(flag, '')
            explanation = explanation[1:] if explanation[0] in PUNCTUATION else explanation
            explanation = explanation.strip().replace('given', 'related')
        
            return {'idx': idx, 'question': question, 'answer': answer, 'explanation': explanation,
                    'ctx_content': chunk_contents, 'chunk_titles': chunk_titles}
        except:
            return {}