tydiqa_gen_978d2a.py

from opencompass.openicl.icl_prompt_template import PromptTemplate
from opencompass.openicl.icl_retriever import ZeroRetriever
from opencompass.openicl.icl_inferencer import GenInferencer
from opencompass.datasets import TydiQADataset, TydiQAEvaluator

# All configs are for TydiQA Goldp task
tydiqa_reader_cfg = dict(
    input_columns=["passage_text", "question_text"],
    output_column="answer"
)

langs = ['arabic', 'bengali', 'english', 'finnish', 'indonesian', 'japanese', 'korean', 'russian', 'swahili', 'telugu', 'thai']

prefixs_prompt = {
    "english": ("Answer the following question based on the information in the given passage.", "Passage:", "Question:", "Answer:"),
    "arabic": ("أجب على السؤال التالي بناءً على المعلومات في المقطع المعطى.", "المقطع:", "السؤال:", "الإجابة:"),
    "bengali": ("প্রদত্ত অধ্যায়ের তথ্যের উপর ভিত্তি করে নিম্নলিখিত প্রশ্নের উত্তর দিন।", "অধ্যায়:", "প্রশ্ন:", "উত্তর:"),
    "finnish": ("Vastaa seuraavaan kysymykseen annetun kappaleen tiedon perusteella.", "Kappale:", "Kysymys:", "Vastaus:"),
    "indonesian": ("Jawab pertanyaan berikut berdasarkan informasi di bagian yang diberikan.", "Bagian:", "Pertanyaan:", "Jawaban:"),
    "korean": ("주어진 문단의 정보에 기반하여 다음 질문에 답하십시오.", "문단:", "질문:", "답변:"),
    "japanese":("文脈に基づいて質問に答えてください。","ぶんしょう:","しつもん:", "かいとう:"),
    "russian": ("Ответьте на следующий вопрос на основе информации в данном отрывке.", "Отрывок:", "Вопрос:", "Ответ:"),
    "swahili": ("Jibu swali lifuatalo kulingana na habari kwenye kifungu kilichotolewa.", "Kifungu:", "Swali:", "Jibu:"),
    "telugu": ("ఇచ్చిన పేరాలోని సమాచారం ఆధారంగా కింది ప్రశ్నకు సమాధానం ఇవ్వండి.", "పేరా:", "ప్రశ్న:", "సమాధానం:"),
    "thai":("ตอบคำถามต่อไปนี้โดยอิงตามข้อมูลในตอนข้อความที่กำหนด:", "ตอนข้อความ:", "คำถาม:", "คำตอบ:")
}

tydiqa_datasets = []
for _lang in langs:
    _hint = prefixs_prompt[_lang]
    tydiqa_infer_cfg = dict(
        prompt_template=dict(
            type=PromptTemplate,
            template=f"{_hint[0]}\n\n</E>{_hint[1]}{{passage_text}}\n{_hint[2]} {{question_text}}\n{_hint[3]} {{answer}}" ,
            ice_token='</E>'
        ),
        retriever=dict(type=ZeroRetriever),
        inferencer=dict(type=GenInferencer), max_out_len=50
    )

    tydiqa_eval_cfg = dict(
        evaluator=dict(type=TydiQAEvaluator),
        ds_split='validation',
        ds_column='answer',
    )

    tydiqa_datasets.append(
        dict(abbr=f'tyidqa-goldp_{_lang}',
            type=TydiQADataset,
            path='./data/tydiqa',
            lang=_lang,
            reader_cfg=tydiqa_reader_cfg,
            infer_cfg=tydiqa_infer_cfg,
            eval_cfg=tydiqa_eval_cfg
        )
    )