China

Large language models (LLMs) enable long-context tasks but face efficiency challenges due to the growing key-value (KV) cache. We propose LeanK, a learning-based method that prunes unimportant key (K) cache channels by leveraging static channel sparsity. LeanK reduces GPU memory and accelerates decoding without sacrificing accuracy. Experiments demonstrate up to 70% K cache and 16%–18% V cache memory reduction, and 1.45× decoding speedup. We also provide insights into model channels and attention heads during long-context inference by analyzing the learned importance distribution. Our code is anonymously available at https://anonymous.4open.science/r/LeanK-7A87/README.md.

EMNLP 2025

LeanK: Learnable K Cache Channel Pruning for Efficient Decoding

llm efficiency

kv cache

pruning

poster

## Welcome!
"I am excited to welcome you to this year’s edition of the Conference on Empirical Methods in Natural Language Processing! Importantly, it marks the 30th edition of EMNLP. With over 8,000 submissions, more than 3,000 accepted papers, and thousands of attendees, we have come a long way from that first
workshop, which had 14 accepted papers. As the field looks ahead, Suzhou is the fitting location for celebrating this milestone: rooted in a long literary tradition, yet modern and forward-looking, and home to a large share of the NLP community."<br>

*Message from the General Chair, Dirk Hovy*

[**Link to Conference Handbook**](https://drive.google.com/file/d/1johU5QqVVYO4RfH7QcIORr7qrVBdzdwC/view?usp=sharing)





<br>

Celebrate 30 Years of EMNLP! 
EMNLP 2025 will be held in Suzhou, China from November 5th to November 9th, 2025.

Mixture of Experts (MoE) architectures have emerged as a promising paradigm for scaling model capacity through top-k routing mechanisms. Although reducing the number of activated experts inherently enables inference acceleration, this efficiency gain typically comes at the cost of significant performance degradation. To address this trade-off between efficiency and performance, we propose HookMoE, a plug-and-play single-layer compensation framework that effectively restores performance using only a small post-training calibration set. Our method strategically inserts a lightweight trainable Hook module immediately preceding selected transformer blocks. Comprehensive evaluations on four popular MoE models, with an average performance degradation of only 2.5% across various benchmarks, our method reduces the number of activated experts by more than 50% and achieves a 1.42times inference speed-up during the prefill stage. Through systematic analysis, we further reveal that the upper layers require fewer active experts, offering actionable insights for refining dynamic expert selection strategies and enhancing the overall efficiency of MoE models.

HookMoE: A learnable performance compensation strategy of Mixture-of-Experts for LLM inference acceleration

Natural Language Inference (NLI) is a fundamental task in natural language processing. While NLI has developed many sub-directions such as sentence-level NLI, document-level NLI and cross-lingual NLI, Cross-Document Cross-Lingual NLI (CDCL-NLI) remains largely unexplored. In this paper, we propose a novel paradigm: CDCL-NLI, which extends traditional NLI capabilities to multi-document, multilingual scenarios. To support this task, we construct a high-quality CDCL-NLI dataset including 25,410 instances and spanning 26 languages. To address the limitations of previous methods on CDCL-NLI task, we further propose an innovative method that integrates RST-enhanced graph fusion with interpretability-aware prediction. Our approach leverages RST (Rhetorical Structure Theory) within heterogeneous graph neural networks for cross-document context modeling, and employs a structure-aware semantic alignment based on lexical chains for cross-lingual understanding. For NLI interpretability, we develop an EDU (Elementary Discourse Unit)-level attribution framework that produces extractive explanations. Extensive experiments demonstrate our approach's superior performance, achieving significant improvements over both conventional NLI models as well as large language models. Our work sheds light on the study of NLI and will bring research interest on cross-document cross-lingual context understanding, hallucination elimination and interpretability inference. Our dataset and code are available at https://anonymous.4open.science/r/CDCL-NLI-637E/ for peer review.

Cross-Document Cross-Lingual NLI via RST-Enhanced Graph Fusion and Interpretability Prediction

Orthogonal finetuning (OFT) offers highly parameter-efficient adaptation while preventing catastrophic forgetting, but its high runtime and memory demands limit practical deployment. We identify the core computational bottleneck in OFT as its weight-centric implementation, which relies on costly matrix-matrix multiplications with cubic complexity. To overcome this, we propose OFTv2, an input-centric reformulation that instead uses matrix-vector multiplications (i.e., matrix-free computation), reducing the computational cost to quadratic. We further introduce the Cayley–Neumann parameterization, an efficient orthogonal parameterization that approximates the matrix inversion in Cayley transform via a truncated Neumann series. These modifications allow OFTv2 to achieve up to 10x faster training and 3x lower GPU memory usage without compromising performance. In addition, we extend OFTv2 to support finetuning quantized foundation models and show that it outperforms the popular QLoRA in training stability, efficiency, and memory usage.

Orthogonal Finetuning Made Scalable

Large language models (LLMs) rely on key-value cache (KV cache) to accelerate decoding by reducing redundant computations. However, the KV cache memory usage grows substantially with longer text sequences, posing challenges for efficient deployment. Existing KV cache eviction methods prune tokens using prefilling-stage attention scores, causing inconsistency with actual inference queries, especially under tight memory budgets. In this paper, we propose Lookahead Q-Cache (LAQ), a novel eviction framework that generates low-cost pseudo lookahead queries to better approximate the true decoding-stage queries. By using these lookahead queries as the observation window for importance estimation, LAQ achieves more consistent and accurate KV cache eviction aligned with real inference scenarios. Experimental results on LongBench and Needle-in-a-Haystack benchmarks show that LAQ outperforms existing methods across various budget levels, achieving a 1 ~ 4 point improvement on LongBench under limited cache budget. Moreover, LAQ is complementary to existing approaches and can be flexibly combined to yield further improvements.

Lookahead Q-Cache: Achieving More Consistent KV Cache Eviction via Pseudo Query

Prompt transfer is a transfer learning method based on prompt tuning, which enhances the parameter performance of prompts in target tasks by transferring source prompt embeddings. Among existing methods, weighted aggregation is effective and possesses the advantages of being lightweight and modular. However, these methods may transfer redundant or irrelevant information from the source prompts to the target prompt, leading to negative impacts. To alleviate this problem, we propose Prompt Contrastive Transformation (PCT), which achieves efficient prompt transfer through prompt contrastive transformation and attentional fusion. PCT transforms the source prompt into task-agnostic embedding and task-specific embeddings through singular value decomposition and contrastive learning, reducing information redundancy among source prompts. The attention module in PCT selects more effective task-specific embeddings and fuses them with task-agnostic embedding into the target prompt. Experimental results show that, despite tuning only 0.035% of task-specific parameters, PCT achieves improvements in prompt transfer for single target task adaptation across various NLP tasks.

Prompt Contrastive Transformation: An Enhanced Strategy for Efficient Prompt Transfer in Natural Language Processing

Powerful large language models (LLMs) are increasingly expected to be deployed with lower computational costs, enabling their capabilities on resource-constrained devices. Post-training quantization (PTQ) has emerged as a star approach to achieve this ambition, with best methods compressing weights to less than 2 bit on average. In this paper, we propose Channel-Relaxed Vector Quantization (CRVQ), a novel technique that significantly improves the performance of PTQ baselines at the cost of only minimal additional bits. This state-of-the-art extreme compression method achieves its results through two key innovations: (1) carefully selecting and reordering a very small subset of critical weight channels, and (2) leveraging extended codebooks to relax the constraint of critical channels. With our method, we demonstrate a 38.9% improvement over the current strongest sub-2-bit PTQ baseline, enabling nearer lossless 1-bit compression. Furthermore, our approach offers flexible customization of quantization bit-width and performance, providing a wider range of deployment options for diverse hardware platforms. Code and checkpoints are available at https://github.com/xuyuzhuang11/CRVQ.

CRVQ: Channel-Relaxed Vector Quantization for Extreme Compression of LLMs

In this paper, we introduce EasyEdit2, a framework designed to enable plug-and-play adjustability for controlling Large Language Model (LLM) behaviors. EasyEdit2 supports a wide range of test-time interventions, including safety, sentiment, personality, reasoning patterns, factuality, and language features. Unlike its predecessor, EasyEdit2 features a new architecture specifically designed for seamless model steering. It comprises key modules such as the steering vector generator and the steering vector applier, which enable automatic generation and application of steering vectors to influence the model’s behavior without modifying its parameters. One of the main advantages of EasyEdit2 is its ease of use—users do not need extensive technical knowledge. With just a single example, they can effectively guide and adjust the model’s responses, making precise control both accessible and efficient. Empirically, we report model steering performance across different LLMs, demonstrating the effectiveness of these techniques. We have released the source code on https://github.com/zjunlp/EasyEdit along with a demonstration notebook. In addition, we provide an online system at http://easyedit.zjukg.cn/ for real-time model steering, and a demo video at https://www.youtube.com/watch?v=AkfoiPfp5rQ.

EasyEdit2: An Easy-to-use Steering Framework for Editing Large Language Models

In this work, we present PoliCorp, a web portal designed to facilitate the search and analysis of political text corpora. PoliCorp provides researchers with access to rich textual data, enabling in-depth analysis of parliamentary discourse over time. The platform currently contains a collection of transcripts of debates from the German parliament, spanning 76 years of proceedings. With the advanced search functionality, researchers can apply logical operations to combine or exclude search criteria, making it easier to filter through vast amounts of parliamentary debate data. The search can be customised by combining multiple fields and applying logical operators to uncover complex patterns and insights within the data. Additional data processing steps were performed to enable web-based search and incorporate supplementary features. A key feature that differentiates PoliCorp is its intuitive web-based interface that enables users to query processed political texts without requiring programming skills. The user-friendly platform allows the creation of custom subcorpora via search parameters, which can be freely downloaded in JSON format for further analysis.

Open Political Corpora: Structuring, Searching, and Analyzing Political Text Collections with PoliCorp

Modern moderation systems increasingly support multiple languages, but often fail to address localisation and low-resource variants—creating safety gaps in real-world deployments. Small models offer a potential alternative to large LLMs, yet still demand considerable data and compute. We present LionGuard 2, a lightweight, multilingual moderation classifier tailored to the Singapore context, supporting English, Chinese, Malay, and partial Tamil. Built on pre-trained OpenAI embeddings and a multi-head ordinal classifier, LionGuard 2 outperforms several commercial and open-source systems across 17 benchmarks, including both Singapore-specific and public English datasets. The system is actively deployed within the Singapore Government, demonstrating practical efficacy at scale. Our findings show that high-quality local data and robust multilingual embeddings can achieve strong moderation performance, without fine-tuning large models. We release our model weights and part of our training data to support future work on LLM safety.

LionGuard 2: Building Lightweight, Data-Efficient & Localised Multilingual Content Moderators

BRAT is a widely used web-based text annotation tool. However, it lacks robust Python support for effective annotation management and processing. We present Bratly, an open-source extension of BRAT that introduces a solid Python backend, enabling advanced functionalities such as annotation typings, collection typings with statistical insights, corpus and annotation handling, object modifications, and entity-level evaluation based on MUC-5 standards. These enhancements streamline annotation workflows, improve usability, and facilitate high-quality NLP research. This paper outlines the system's architecture, functionalities and evaluation, positioning it as a valuable BRAT extension for its users. The tool is open-source, and the NLP community is welcome to suggest improvements.

Downloads

Next from EMNLP 2025

HookMoE: A learnable performance compensation strategy of Mixture-of-Experts for LLM inference acceleration