README.md

metadata

task_categories:
  - question-answering
  - text-generation
  - summarization
  - conversational
  - text-classification
language:
  - en
  - zh
tags:
  - Long Context
size_categories:
  - 1K<n<10K

Introduction

LongBench is the first comprehensive dataset for multi-language, multi-task, and comprehensive assessment of long text understanding capabilities of large language models. In the context of the widespread attention to the multi-language capabilities of large models, LongBench includes different languages (Chinese and English) to provide a more comprehensive evaluation of the large models' multi-language capabilities in long texts. In addition, LongBench consists of twenty different tasks, covering key long-text application scenarios such as single-document QA, multi-document QA, summaries, few-shot learning, code completion, and synthesis tasks.

We are fully aware of the potentially high costs involved in the model evaluation process, especially in the context of long-text scenarios (such as manual annotation costs or API call costs). Therefore, we have adopted a fully automated evaluation method, aimed at measuring and evaluating the model's ability to understand long texts at the lowest cost and most effectively.

LongBench includes 13 English tasks, 5 Chinese tasks, and 2 code tasks, with the average length of most tasks ranging from 5k to 15k. From the main task categories, LongBench includes six types of tasks, namely multi-document QA, single-document QA, summaries, Few-shot learning, synthetic tasks, and code completion.

How to use it?

Loading Data

from datasets import load_dataset

datasets = ["hotpotqa", "2wikimqa", "musique", "dureader", "narrativeqa", "qasper", "multifieldqa_en", \
    "multifieldqa_zh", "gov_report", "qmsum", "vcsum", "trec", "nq", "triviaqa", "lsht", "passage_count", \
    "passage_retrieval_en", "passage_retrieval_zh", "lcc", "repobench-p"]

for dataset in datasets:
    data = load_dataset('THUDM/LongBench', dataset, split='test')

Data Format

All data in LongBench are standardized to the following format:

{
    "input": "The input/command for the task, usually short, such as questions in QA, queries in Few-shot tasks, etc.",
    "context": "The long context text required for the task, such as documents, cross-file code, few-shot samples in Few-shot tasks",
    "answers": "List composed of all standard answers",
    "length": "Total length of the first three items of text (counted in characters for Chinese and words for English)",
    "dataset": "The name of the dataset to which this piece of data belongs",
    "language": "The language of this piece of data",
    "all_classes": "All categories in classification tasks, null for non-classification tasks",
    "_id": "Random id for each piece of data"
}

Evaluation

This repository provides data download for LongBench. If you wish to use this dataset for automated evaluation, please refer to our github.

Task statistics

Task	Task Type	Eval metric	Avg len	Language	#Sample
HotpotQA	Multi-doc QA	F1	9149	EN	200
2WikiMultihopQA	Multi-doc QA	F1	4885	EN	200
Musique	Multi-doc QA	F1	7798	EN	200
DuReader	Multi-doc QA	Rouge-L	15768	ZH	200
MultiFieldQA-en	Single-doc QA	F1	4559	EN	150
MultiFieldQA-zh	Single-doc QA	F1	6771	ZH	200
NarrativeQA	Single-doc QA	F1	18405	EN	200
Qasper	Single-doc QA	F1	3619	EN	200
GovReport	Summarization	Rouge-L	8169	EN	200
QMSum	Summarization	Rouge-L	10546	EN	200
VCSUM	Summarization	Rouge-L	15147	ZH	200
TriviaQA	Few shot	F1	8015	EN	200
NQ	Few shot	F1	8210	EN	200
TREC	Few shot	Accuracy	5176	EN	200
LSHT	Few shot	Accuracy	22333	ZH	200
PassageRetrieval-en	Synthetic	Accuracy	9288	EN	200
PassageCount	Synthetic	Accuracy	11141	EN	200
PassageRetrieval-zh	Synthetic	Accuracy	6745	ZH	200
LCC	Code	Edit Sim	1235	Python/C#/Java	500
RepoBench-P	Code	Edit Sim	5622	Python/Java	500

Note: In order to avoid discrepancies caused by different tokenizers, we use the word count (using Python's split function) to calculate the average length of English datasets and code datasets, and use the character count to calculate the average length of Chinese datasets.

Task description

Task	Task Description
HotpotQA	Answer related questions based on multiple given documents
2WikiMultihopQA	Answer related questions based on multiple given documents
Musique	Answer related questions based on multiple given documents
DuReader	Answer related Chinese questions based on multiple retrieved documents
MultiFieldQA-en	Answer English questions based on a single document, which comes from a relatively diverse field
MultiFieldQA-zh	Answer Chinese questions based on a single document, which comes from a relatively diverse field
NarrativeQA	Ask questions based on stories or scripts, including understanding of important elements such as characters, plots, themes, etc.
Qasper	Ask questions based on a single paper, questions proposed by NLP readers, and answered by NLP practitioners
GovReport	A summarization task that requires summarizing government work reports
QMSum	A summarization task that requires summarizing meeting records based on user queries
VCSUM	A summarization task that requires summarizing Chinese meeting records
TriviaQA	Single document question answering task, providing several Few Shot examples
NQ	Single document question answering task, providing several Few Shot examples
TREC	A classification task that requires categorizing questions, includes 50 categories in total
LSHT	A Chinese classification task that requires categorizing news, includes 24 categories in total
PassageRetrieval-en	Given 30 English Wikipedia paragraphs, determine which paragraph the given summary belongs to
PassageCount	Determine the number of non-repeating paragraphs in a given number of paragraphs
PassageRetrieval-zh	Given several Chinese paragraphs from the C4 data set, determine which paragraph the given abstract belongs to
LCC	Given a longer piece of code, predict the next line of code
RepoBench-P	Given code in multiple files within a GitHub repository (including inter-file dependencies), predict the next line of code

Task construction

Note: For all tasks constructed from existing datasets, we use data from the validation or test set of the existing dataset (except for VCSUM).

• The tasks of HotpotQA, 2WikiMultihopQA, Musique, and DuReader are built based on the original datasets and processed to make them suitable for long text evaluation. Specifically, for questions in the validation set, we select the evidence passage that contains the answer and several distracting articles. These articles together with the original question constitute the input of the related tasks.
• The tasks of MultiFiedQA-zh and MultiFieldQA-en consist of long-text data from about 10 sources, including Latex papers, judicial documents, government work reports, and PDF documents indexed by Google. For each long text, we invite several PhD and master students to annotate, i.e., to ask questions based on the long text and give the correct answers. To better automate evaluation, we require the annotators to ask questions with definitive answers as much as possible.
• The tasks of NarrativeQA, Qasper, GovReport, and QMSum directly use the data provided by the original papers. In the specific construction, we use the template provided by ZeroSCROLLS to convert the corresponding data into pure text input.
• The VCSUM task is built based on the original dataset, and we have designed a corresponding template to convert the corresponding data into pure text input.
• The tasks of TriviaQA and NQ are constructed in the manner of CoLT5, which provides several examples of question and answering based on documents, and requires the language model to answer related questions based on new documents.
• The tasks of TREC and LSHT are built based on the original datasets. For each question in the validation set, we sample several data from the training set to form few-shot examples. These examples together with the questions in the validation set constitute the input for this task.
• The PassageRetrieval-en task is constructed based on English Wikipedia. For each piece of data, we randomly sample 30 paragraphs from English Wikipedia and select one for summarization (using GPT3.5 Turbo). The task requires the model to specify which original paragraph the summary corresponds to.
• The PassageCount task is constructed based on the English wiki. For each piece of data, we randomly sample several passages from English Wikipedia, repeat each paragraph at random several times, and finally shuffle the paragraphs at random. This task requires the model to determine how many paragraphs are not repeated in a given number of paragraphs.
• The PasskeyRetrieval-zh task is constructed based on C4. For each piece of data, we randomly sample several Chinese paragraphs from C4 and select one of them for summary (using GPT3.5 Turbo). This task requires the model to give the original paragraph to which the summary should correspond.
• For the LCC task, we sample from the original code completion dataset. In the RepoBench-P task, we select the most challenging XF-F (Cross-File-First) setting from the original dataset and refer to the Oracle-Filled scenario in the original text. For each original piece of data, we randomly extract multiple cross-file code snippets, including the effective cross-file code snippet (gold snippet), and concatenate them as input, requiring the model to use the effective cross-file code to complete the current file's code.