Don't Make Your LLM an Evaluation Benchmark Cheater

Large language models~(LLMs) have greatly advanced the frontiers of artificial intelligence, attaining remarkable improvement in model capacity. To assess the model performance, a typical approach is to construct evaluation benchmarks for measuring the ability level of LLMs in different aspects. Despite that a number of high-quality benchmarks have been released, the concerns about the appropriate use of these benchmarks and the fair comparison of different models are increasingly growing. Considering these concerns, in this paper, we discuss the potential risk and impact of inappropriately using evaluation benchmarks and misleadingly interpreting the evaluation results. Specially, we focus on a special issue that would lead to inappropriate evaluation, \ie \emph{benchmark leakage}, referring that the data related to evaluation sets is occasionally used for model training. This phenomenon now becomes more common since pre-training data is often prepared ahead of model test. We conduct extensive experiments to study the effect of benchmark leverage, and find that it can dramatically boost the evaluation results, which would finally lead to an unreliable assessment of model performance. To improve the use of existing evaluation benchmarks, we finally present several guidelines for both LLM developers and benchmark maintainers. We hope this work can draw attention to appropriate training and evaluation of LLMs.

Instruct and Extract: Instruction Tuning for On-Demand Information Extraction

Large language models with instruction-following capabilities open the door to a wider group of users. However, when it comes to information extraction - a classic task in natural language processing - most task-specific systems cannot align well with long-tail ad hoc extraction use cases for non-expert users. To address this, we propose a novel paradigm, termed On-Demand Information Extraction, to fulfill the personalized demands of real-world users. Our task aims to follow the instructions to extract the desired content from the associated text and present it in a structured tabular format. The table headers can either be user-specified or inferred contextually by the model. To facilitate research in this emerging area, we present a benchmark named InstructIE, inclusive of both automatically generated training data, as well as the human-annotated test set. Building on InstructIE, we further develop an On-Demand Information Extractor, ODIE. Comprehensive evaluations on our benchmark reveal that ODIE substantially outperforms the existing open-source models of similar size. Our code and dataset are released on https://github.com/yzjiao/On-Demand-IE.

"Why Should I Review This Paper?" Unifying Semantic, Topic, and Citation Factors for Paper-Reviewer Matching

As many academic conferences are overwhelmed by a rapidly increasing number of paper submissions, automatically finding appropriate reviewers for each submission becomes a more urgent need than ever. Various factors have been considered by previous attempts on this task to measure the expertise relevance between a paper and a reviewer, including whether the paper is semantically close to, shares topics with, and cites previous papers of the reviewer. However, the majority of previous studies take only one of these factors into account, leading to an incomprehensive evaluation of paper-reviewer relevance. To bridge this gap, in this paper, we propose a unified model for paper-reviewer matching that jointly captures semantic, topic, and citation factors. In the unified model, a contextualized language model backbone is shared by all factors to learn common knowledge, while instruction tuning is introduced to characterize the uniqueness of each factor by producing factor-aware paper embeddings. Experiments on four datasets (one of which is newly contributed by us) across different fields, including machine learning, computer vision, information retrieval, and data mining, consistently validate the effectiveness of our proposed UniPR model in comparison with state-of-the-art paper-reviewer matching methods and scientific pre-trained language models.

The Shifted and The Overlooked: A Task-oriented Investigation of User-GPT Interactions

Recent progress in Large Language Models (LLMs) has produced models that exhibit remarkable performance across a variety of NLP tasks. However, it remains unclear whether the existing focus of NLP research accurately captures the genuine requirements of human users. This paper provides a comprehensive analysis of the divergence between current NLP research and the needs of real-world NLP applications via a large-scale collection of user-GPT conversations. We analyze a large-scale collection of real user queries to GPT. We compare these queries against existing NLP benchmark tasks and identify a significant gap between the tasks that users frequently request from LLMs and the tasks that are commonly studied in academic research. For example, we find that tasks such as ``design'' and ``planning'' are prevalent in user interactions but are largely neglected or different from traditional NLP benchmarks. We investigate these overlooked tasks, dissect the practical challenges they pose, and provide insights toward a roadmap to make LLMs better aligned with user needs.

Seeking Neural Nuggets: Knowledge Transfer in Large Language Models from a Parametric Perspective

Large Language Models (LLMs) inherently encode a wealth of knowledge within their parameters through pre-training on extensive corpora. While prior research has delved into operations on these parameters to manipulate the underlying implicit knowledge (encompassing detection, editing, and merging), there remains an ambiguous understanding regarding their transferability across models with varying scales. In this paper, we seek to empirically investigate knowledge transfer from larger to smaller models through a parametric perspective. To achieve this, we employ sensitivity-based techniques to extract and align knowledge-specific parameters between different LLMs. Moreover, the LoRA module is used as the intermediary mechanism for injecting the extracted knowledge into smaller models. Evaluations across four benchmarks validate the efficacy of our proposed method. Our findings highlight the critical factors contributing to the process of parametric knowledge transfer, underscoring the transferability of model parameters across LLMs of different scales. We release code and data at \url{https://github.com/maszhongming/ParaKnowTransfer}.

Language Models As Semantic Indexers

Semantic identifier (ID) is an important concept in information retrieval that aims to preserve the semantics of objects such as documents and items inside their IDs. Previous studies typically adopt a two-stage pipeline to learn semantic IDs by first procuring embeddings using off-the-shelf text encoders and then deriving IDs based on the embeddings. However, each step introduces potential information loss and there is usually an inherent mismatch between the distribution of embeddings within the latent space produced by text encoders and the anticipated distribution required for semantic indexing. Nevertheless, it is non-trivial to design a method that can learn the document's semantic representations and its hierarchical structure simultaneously, given that semantic IDs are discrete and sequentially structured, and the semantic supervision is deficient. In this paper, we introduce LMINDEXER, a self-supervised framework to learn semantic IDs with a generative language model. We tackle the challenge of sequential discrete ID by introducing a semantic indexer capable of generating neural sequential discrete representations with progressive training and contrastive learning. In response to the semantic supervision deficiency, we propose to train the model with a self-supervised document reconstruction objective. The learned semantic indexer can facilitate various downstream tasks, such as recommendation and retrieval. We conduct experiments on three tasks including recommendation, product search, and document retrieval on five datasets from various domains, where LMINDEXER outperforms competitive baselines significantly and consistently.

Ontology Enrichment for Effective Fine-grained Entity Typing

Fine-grained entity typing (FET) is the task of identifying specific entity types at a fine-grained level for entity mentions based on their contextual information. Conventional methods for FET require extensive human annotation, which is time-consuming and costly. Recent studies have been developing weakly supervised or zero-shot approaches. We study the setting of zero-shot FET where only an ontology is provided. However, most existing ontology structures lack rich supporting information and even contain ambiguous relations, making them ineffective in guiding FET. Recently developed language models, though promising in various few-shot and zero-shot NLP tasks, may face challenges in zero-shot FET due to their lack of interaction with task-specific ontology. In this study, we propose OnEFET, where we (1) enrich each node in the ontology structure with two types of extra information: instance information for training sample augmentation and topic information to relate types to contexts, and (2) develop a coarse-to-fine typing algorithm that exploits the enriched information by training an entailment model with contrasting topics and instance-based augmented training samples. Our experiments show that OnEFET achieves high-quality fine-grained entity typing without human annotation, outperforming existing zero-shot methods by a large margin and rivaling supervised methods.

Learning Multiplex Embeddings on Text-rich Networks with One Text Encoder

In real-world scenarios, texts in a network are often linked by multiple semantic relations (e.g., papers in an academic network are referenced by other publications, written by the same author, or published in the same venue), where text documents and their relations form a multiplex text-rich network. Mainstream text representation learning methods use pretrained language models (PLMs) to generate one embedding for each text unit, expecting that all types of relations between texts can be captured by these single-view embeddings. However, this presumption does not hold particularly in multiplex text-rich networks. Along another line of work, multiplex graph neural networks (GNNs) directly initialize node attributes as a feature vector for node representation learning, but they cannot fully capture the semantics of the nodes' associated texts. To bridge these gaps, we propose METERN, a new framework for learning Multiplex Embeddings on TExt-Rich Networks. In contrast to existing methods, METERN uses one text encoder to model the shared knowledge across relations and leverages a small number of parameters per relation to derive relation-specific representations. This allows the encoder to effectively capture the multiplex structures in the network while also preserving parameter efficiency. We conduct experiments on nine downstream tasks in five networks from both academic and e-commerce domains, where METERN outperforms baselines significantly and consistently. The code is available at https://github.com/PeterGriffinJin/METERN-submit.

Model Tells You What to Discard: Adaptive KV Cache Compression for LLMs

In this study, we introduce adaptive KV cache compression, a plug-and-play method that reduces the memory footprint of generative inference for Large Language Models (LLMs). Different from the conventional KV cache that retains key and value vectors for all context tokens, we conduct targeted profiling to discern the intrinsic structure of attention modules. Based on the recognized structure, we then construct the KV cache in an adaptive manner: evicting long-range contexts on attention heads emphasizing local contexts, discarding non-special tokens on attention heads centered on special tokens, and only employing the standard KV cache for attention heads that broadly attend to all tokens. Moreover, with the lightweight attention profiling used to guide the construction of the adaptive KV cache, FastGen can be deployed without resource-intensive fine-tuning or re-training. In our experiments across various asks, FastGen demonstrates substantial reduction on GPU memory consumption with negligible generation quality loss. We will release our code and the compatible CUDA kernel for reproducibility.