[Feature] Add the implement of QuALITY datasets (#976)

#976

configs/datasets/QuALITY/QuALITY.md

@@ -0,0 +1,56 @@
+# QuALITY
+## Introduction
+The following introduction comes from the description in [QuALITY Leaderboard](https://nyu-mll.github.io/quality/)
+
+```
+QuALITY is a multiple-choice question answering dataset with context passages in English that have an average length of about 5,000 tokens.
+```
+
+These questions were categorized into two levels: easy and hard.
+
+## Official link
+
+### Paper
+
+[QuALITY: Question Answering with Long Input Texts, Yes!](https://arxiv.org/pdf/2112.08608.pdf)
+
+### Repository
+
+[nyu-mll/quality](https://github.com/nyu-mll/quality)
+
+
+## Evaluation results
+
+```
+dataset    version    metric    mode      qwen1.5-7b-chat-hf    qwen1.5-14b-chat-hf    qwen1.5-72b-chat-hf
+---------  ---------  --------  ------  --------------------  ---------------------  ---------------------
+QuALITY    ed2404     easy_acc  gen                    62.39                  68.17                  76.69
+QuALITY    ed2404     hard_acc  gen                    49.27                  56.22                  63.96
+QuALITY    ed2404     all_acc   gen                    54.65                  60.88                  68.84
+```
+
+## Reference
+```
+@inproceedings{pang-etal-2022-quality,
+    title = "{Q}u{ALITY}: Question Answering with Long Input Texts, Yes!",
+    author = "Pang, Richard Yuanzhe  and
+      Parrish, Alicia  and
+      Joshi, Nitish  and
+      Nangia, Nikita  and
+      Phang, Jason  and
+      Chen, Angelica  and
+      Padmakumar, Vishakh  and
+      Ma, Johnny  and
+      Thompson, Jana  and
+      He, He  and
+      Bowman, Samuel",
+    booktitle = "Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies",
+    month = jul,
+    year = "2022",
+    address = "Seattle, United States",
+    publisher = "Association for Computational Linguistics",
+    url = "https://aclanthology.org/2022.naacl-main.391",
+    pages = "5336--5358",
+    abstract = "To enable building and testing models on long-document comprehension, we introduce QuALITY, a multiple-choice QA dataset with context passages in English that have an average length of about 5,000 tokens, much longer than typical current models can process. Unlike in prior work with passages, our questions are written and validated by contributors who have read the entire passage, rather than relying on summaries or excerpts. In addition, only half of the questions are answerable by annotators working under tight time constraints, indicating that skimming and simple search are not enough to consistently perform well. Our baseline models perform poorly on this task (55.4{\%}) and significantly lag behind human performance (93.5{\%}).",
+}
+```

configs/datasets/QuALITY/QuALITY_gen.py

@@ -0,0 +1,4 @@
+from mmengine.config import read_base
+
+with read_base():
+    from .QuALITY_gen_c407cb import QuALITY_datasets  # noqa: F401, F403

configs/datasets/QuALITY/QuALITY_gen_c407cb.py

@@ -0,0 +1,38 @@
+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 QuALITYDataset, QuALITYEvaluator
+from opencompass.utils.text_postprocessors import first_option_postprocess
+
+QuALITY_reader_cfg = dict(
+    input_columns=['article', 'question', 'A', 'B', 'C', 'D'],
+    output_column='gold_label',
+)
+
+QuALITY_infer_cfg = dict(
+    prompt_template=dict(
+        type=PromptTemplate,
+        template=dict(round=[
+            dict(
+                role="HUMAN",
+                prompt=
+                "Read the article, and answer the question.\n\nArticle:\n{article}\n\nQ: {question}\n\nA. {A}\nB. {B}\nC. {C}\nD. {D}"
+            ),
+        ])),
+    retriever=dict(type=ZeroRetriever),
+    inferencer=dict(type=GenInferencer))
+
+QuALITY_eval_cfg = dict(
+    evaluator=dict(type=QuALITYEvaluator),
+    pred_postprocessor=dict(type=first_option_postprocess, options='ABCD'),
+    pred_role='BOT')
+
+QuALITY_datasets = [
+    dict(
+        abbr='QuALITY',
+        type=QuALITYDataset,
+        path='./data/QuALITY/QuALITY.v1.0.1.htmlstripped.dev',
+        reader_cfg=QuALITY_reader_cfg,
+        infer_cfg=QuALITY_infer_cfg,
+        eval_cfg=QuALITY_eval_cfg),
+]

opencompass/datasets/QuALITY.py

@@ -0,0 +1,59 @@
+import json
+
+from datasets import Dataset
+
+from opencompass.openicl.icl_evaluator import BaseEvaluator
+from opencompass.registry import LOAD_DATASET
+
+from .base import BaseDataset
+
+
+@LOAD_DATASET.register_module()
+class QuALITYDataset(BaseDataset):
+
+    @staticmethod
+    def load(path: str):
+        dataset_list = []
+        with open(path, 'r', encoding='utf-8') as f:
+            for line in f:
+                line = json.loads(line)
+                for question in line['questions']:
+                    dataset_list.append({
+                        'article':
+                        line['article'],
+                        'question':
+                        question['question'],
+                        'A':
+                        question['options'][0],
+                        'B':
+                        question['options'][1],
+                        'C':
+                        question['options'][2],
+                        'D':
+                        question['options'][3],
+                        'gold_label':
+                        'ABCD'[question['gold_label'] - 1],
+                        'difficult':
+                        question['difficult']
+                    })
+        return Dataset.from_list(dataset_list)
+
+
+class QuALITYEvaluator(BaseEvaluator):
+
+    def score(self, predictions, references, test_set):
+        assert len(predictions) == len(references)
+        easy, hard, all = [], [], []
+        for pred, refer, test in zip(predictions, references, test_set):
+            if pred == refer:
+                answer = True
+            else:
+                answer = False
+            all.append(answer)
+            if test['difficult'] == 0:
+                easy.append(answer)
+            else:
+                hard.append(answer)
+        return dict(easy_acc=sum(easy) / len(easy) * 100,
+                    hard_acc=sum(hard) / len(easy) * 100,
+                    all_acc=sum(all) / len(all) * 100)

opencompass/datasets/__init__.py

@@ -78,6 +78,7 @@ from .piqa import *  # noqa: F401, F403
 from .py150 import *  # noqa: F401, F403
 from .qasper import *  # noqa: F401, F403
 from .qaspercut import *  # noqa: F401, F403
+from .QuALITY import *  # noqa: F401, F403
 from .race import *  # noqa: F401, F403
 from .realtoxicprompts import *  # noqa: F401, F403
 from .reasonbench import ReasonBenchDataset  # noqa: F401, F403