PulseExploreJournal ClubDebatesTrendingResearchersJournals
Instagram
HomeExploreJournal ClubTrending
Synapse
⌘+K
Synapse
June 25, 20240 citationsOpen Access

LongIns: A Challenging Long-context Instruction-based Exam for LLMs

View Full Paper
SGShawn GavinTZTuney ZhengJLJiaheng Liu

Key Points

Key points are not available for this paper at this time.

Abstract

The long-context capabilities of large language models (LLMs) have been a hot topic in recent years. To evaluate the performance of LLMs in different scenarios, various assessment benchmarks have emerged. However, as most of these benchmarks focus on identifying key information to answer questions, which mainly requires the retrieval ability of LLMs, these benchmarks can partially represent the reasoning performance of LLMs from large amounts of information. Meanwhile, although LLMs often claim to have context windows of 32k, 128k, 200k, or even longer, these benchmarks fail to reveal the actual supported length of these LLMs. To address these issues, we propose the LongIns benchmark dataset, a challenging long-context instruction-based exam for LLMs, which is built based on the existing instruction datasets. Specifically, in our LongIns, we introduce three evaluation settings: Global Instruction & Single Task (GIST), Local Instruction & Single Task (LIST), and Local Instruction & Multiple Tasks (LIMT). Based on LongIns, we perform comprehensive evaluations on existing LLMs and have the following important findings: (1). The top-performing GPT-4 with 128k context length performs poorly on the evaluation context window of 16k in our LongIns. (2). For the multi-hop reasoning ability of many existing LLMs, significant efforts are still needed under short context windows (less than 4k).

Ask AI
Helpful
Bookmark
Share
View Full Paper

Cite This Study

Gavin et al. (2024) studied this question.

synapsesocial.com/papers/68e636c5b6db6435875c8d3ahttps://doi.org/10.48550/arxiv.2406.17588
Ask AI
Helpful
Bookmark
Share
View Full Paper