wckwan/M4LE

Introduction

M4LE is a Multi-ability, Multi-range, Multi-task, bilingual benchmark for long-context evaluation. We categorize long-context understanding into five distinct abilities by considering whether it is required to identify single or multiple spans in long contexts based on explicit or semantic hints. Specifically, these abilities are explicit single-span, semantic single-span, explicit multiple-span, semantic multiple-span, and global. Different from previous long-context benchmarks that simply compile from a set of existing long NLP benchmarks, we introduce an automated method to transform short-sequence tasks into a comprehensive long-sequence scenario encompassing all these capabilities.

M4LE consists of 36 tasks, covering 11 task types and 12 domains. For each task, we construct 200 instances for each context length bucket (1K, 2K, 4K, 6K, 8K, 12K, 16K, 24K, 32K, 64K, 128K). Due to computation and cost constraints, our paper evaluated 11 well-established LLMs on instances up to the 8K context length bucket. For more details, please refer to the paper available at https://arxiv.org/abs/2310.19240. You can also explore the GitHub page at https://github.com/KwanWaiChung/M4LE.

Usage

You can load the dataset by specifying the task name:

from datasets import load_dataset
tasks = [
    "arxiv",
    "bigpatent_global_cls",
    "bigpatent_global_sum",
    "booksum",
    "c3",
    "cepsum",
    "clts+",
    "cnewsum",
    "cnnnews",
    "drcd_explicit-single",
    "drcd_semantic-single",
    "duorc",
    "dureader",
    "hotpotqa",
    "lcsts",
    "marc",
    "mnds-news_explicit-single",
    "mnds-news_explicit-multiple",
    "mnds-news_semantic-multiple",
    "ncls",
    "news-commentary-en2zh",
    "news-commentary-zh2en",
    "news2016",
    "newsqa",
    "nq-open",
    "online-shopping",
    "open-subtitles-en2zh",
    "open-subtitles-zh2en",
    "pubmed",
    "tedtalks-en2zh",
    "tedtalks-zh2en",
    "thucnews_explicit-single",
    "thucnews_explicit-multiple",
    "thucnews_semantic-multiple",
    "triviaqa",
    "wiki2019zh",
    "wikihow",
    "wikitext-103",
    "wow",
]

for task in tasks:
    data = load_dataset('wckwan/M4LE', task, split='test')

Format

Each testing instance follows this format:

{
    "instruction": "<task description>",
    "input": "<task input with one-shot example>",
    "answers": ["<answer1>", "<answer2>"],
    "input_length": <int, number of words in instruction and input separated by space>,
    "total_length": <int, number of words in instruction, input and gold answer separated by space>,
    "length_bucket": <int, the length bucket to which this instance belongs>
}

Tasks

Here is the full list of the tasks with their descriptions. For more details about these tasks, please refer to the paper.

Ability	Task Name	Task Type	Language	Description
Explicit Single	mnds-news_explicit-single	CLS + RET	En	Classify a specified news article.
Explicit Single	thucnews_explicit-single	CLS + RET	Zh	Classify a specified news article.
Explicit Single	newsqa	QA + RET	En	Answer a question based on a specified news article.
Explicit Single	c3	QA + RET	Zh	Answer a multi-choice question based on a textbook extract.
Explicit Single	wow	RET	En	Return the ID of the article related to a specified topic.
Explicit Single	drcd_explicit-single	RET	Zh	Return the ID of the article related to a specified topic.
Explicit Single	cnnnews	SUM + RET	En	Summarize a specified news article.
Explicit Single	cepsum	SUM + RET	Zh	Summarize a specified product description.
Explicit Single	lcsts	SUM + RET	Zh	Summarize a specified news article.
Explicit Single	ncls	SUM + RET	En, Zh	Summarize a specified news article.
Explicit Multiple	mnds-news_explicit-multiple	CLS + RET	En	Return the IDs of all the articles belong to a specified class.
Explicit Multiple	thucnews_explicit-multiple	CLS + RET	Zh	Return the IDs of all the articles belong to a specified class.
Explicit Multiple	marc	CLS + RET	En, Zh	Return the IDs of all the positive product reviews.
Explicit Multiple	online-shopping	CLS + RET	Zh	Return the IDs of all the positive product reviews.
Semantic Single	wikitext-103	NLI + RET	En	Return the ID of the paragraph that continues a query paragraph.
Semantic Single	wiki2019zh	NLI + RET	Zh	Return the ID of the paragraph that continues a query paragraph.
Semantic Single	duorc	QA	En	Answer a question based on multiple movie plots.
Semantic Single	nq-open	QA	En	Answer a question based on multiple wikipedia paragraphs.
Semantic Single	dureader	QA	Zh	Answer a question based on multiple web snippets.
Semantic Single	drcd_semantic-single	QA	Zh	Answer a question based on multiple wikipedia paragraphs.
Semantic Single	wikihow	SUM + RET	En	Summarize an article based on a given topic.
Semantic Single	news2016	SUM + RET	Zh	Summarize a news article based on a given title.
Semantic Single	tedtalks-en2zh/tedtalks-zh2en	TRAN + RET	En, Zh	Translate a Ted Talk transcript based on a given title.
Semantic Multiple	mnds-news_semantic-multiple	CLS + CNT	En	Return the number of news articles belonging to a specified class.
Semantic Multiple	thucnews_semantic-multiple	CLS + CNT	Zh	Return the number of news articles belonging to a specified class.
Semantic Multiple	hotpotqa	QA	En	Answer a question based on multiple wikipedia paragraphs.
Global	bigpatent_global_cls	CLS	En	Classify a patent document.
Global	triviaqa	QA	En	Answer a question based on a web snippet.
Global	arxiv	SUM	En	Summarize an academic paper.
Global	bigpatent_global_sum	SUM	En	Summarize a patent document.
Global	pubmed	SUM	En	Summarize a medical paper.
Global	booksum	SUM	En	Summarize one or more chapters of a book.
Global	cnewsum	SUM	Zh	Summarize a news article.
Global	clts+	SUM	Zh	Summarize a news article.
Global	open-subtitles-en2zh/open-subtitles-zh2en	TRAN	En, Zh	Translate the movie subtitles.
Global	news-commentary-en2zh/news-commentary-zh2en	TRAN	En, Zh	Translate the movie subtitles.

Citation

If you find our paper and resources useful, please consider citing our paper:

@misc{kwan_m4le_2023,
  title = {{{M4LE}}: {{A Multi-Ability Multi-Range Multi-Task Multi-Domain Long-Context Evaluation Benchmark}} for {{Large Language Models}}},
  author = {Kwan, Wai-Chung and Zeng, Xingshan and Wang, Yufei and Sun, Yusen and Li, Liangyou and Shang, Lifeng and Liu, Qun and Wong, Kam-Fai},
  year = {2023},
}