PyTorchKR Papers

최신 논문

총 12,491편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2412.19442공식 코드 · Papers with Code

A Survey on Large Language Model Acceleration based on KV Cache Management

Large Language Models (LLMs) have revolutionized a wide range of domains such as natural language processing, computer vision, and multi-modal tasks due to their ability to comprehend context and perform logical reasoning. However, the computational and memory demands of LLMs, particularly during inference, pose significant challenges when scaling them to real-world, long-context, and real-time applications. Key-Value (KV) cache management has emerged as a critical optimization technique for accelerating LLM inference by reducing redundant computations and improving memory utilization. This survey provides a comprehensive overview of KV cache management strategies for LLM acceleration, categorizing them into token-level, model-level, and system-level optimizations. Token-level strategies include KV cache selection, budget allocation, merging, quantization, and low-rank decomposition, while model-level optimizations focus on architectural innovations and attention mechanisms to enhance KV reuse. System-level approaches address memory management, scheduling, and hardware-aware designs to improve efficiency across diverse computing environments. Additionally, the survey provides an overview of both text and multimodal datasets and benchmarks used to evaluate these strategies. By presenting detailed taxonomies and comparative analyses, this work aims to offer useful insights for researchers and practitioners to support the development of efficient and scalable KV cache management techniques, contributing to the practical deployment of LLMs in real-world applications. The curated paper list for KV cache management is in: \href{https://github.com/TreeAI-Lab/Awesome-KV-Cache-Management}{https://github.com/TreeAI-Lab/Awesome-KV-Cache-Management}.

arXiv:2412.19437공식 코드 · Papers with Code

DeepSeek-V3 Technical Report

We present DeepSeek-V3, a strong Mixture-of-Experts (MoE) language model with 671B total parameters with 37B activated for each token. To achieve efficient inference and cost-effective training, DeepSeek-V3 adopts Multi-head Latent Attention (MLA) and DeepSeekMoE architectures, which were thoroughly validated in DeepSeek-V2. Furthermore, DeepSeek-V3 pioneers an auxiliary-loss-free strategy for load balancing and sets a multi-token prediction training objective for stronger performance. We pre-train DeepSeek-V3 on 14.8 trillion diverse and high-quality tokens, followed by Supervised Fine-Tuning and Reinforcement Learning stages to fully harness its capabilities. Comprehensive evaluations reveal that DeepSeek-V3 outperforms other open-source models and achieves performance comparable to leading closed-source models. Despite its excellent performance, DeepSeek-V3 requires only 2.788M H800 GPU hours for its full training. In addition, its training process is remarkably stable. Throughout the entire training process, we did not experience any irrecoverable loss spikes or perform any rollbacks. The model checkpoints are available at https://github.com/deepseek-ai/DeepSeek-V3.

arXiv:2412.19312공식 코드 · Papers with Code

From Interests to Insights: An LLM Approach to Course Recommendations Using Natural Language Queries

Most universities in the United States encourage their students to explore academic areas before declaring a major and to acquire academic breadth by satisfying a variety of requirements. Each term, students must choose among many thousands of offerings, spanning dozens of subject areas, a handful of courses to take. The curricular environment is also dynamic, and poor communication and search functions on campus can limit a student's ability to discover new courses of interest. To support both students and their advisers in such a setting, we explore a novel Large Language Model (LLM) course recommendation system that applies a Retrieval Augmented Generation (RAG) method to the corpus of course descriptions. The system first generates an 'ideal' course description based on the user's query. This description is converted into a search vector using embeddings, which is then used to find actual courses with similar content by comparing embedding similarities. We describe the method and assess the quality and fairness of some example prompts. Steps to deploy a pilot system on campus are discussed.

arXiv:2412.19198

Multi-Attribute Constraint Satisfaction via Language Model Rewriting

Obeying precise constraints on top of multiple external attributes is a common computational problem underlying seemingly different domains, from controlled text generation to protein engineering. Existing language model (LM) controllability methods for multi-attribute constraint satisfaction often rely on specialized architectures or gradient-based classifiers, limiting their flexibility to work with arbitrary black-box evaluators and pretrained models. Current general-purpose large language models, while capable, cannot achieve fine-grained multi-attribute control over external attributes. Thus, we create Multi-Attribute Constraint Satisfaction (MACS), a generalized method capable of finetuning language models on any sequential domain to satisfy user-specified constraints on multiple external real-value attributes. Our method trains LMs as editors by sampling diverse multi-attribute edit pairs from an initial set of paraphrased outputs. During inference, LM iteratively improves upon its previous solution to satisfy constraints for all attributes by leveraging our designed constraint satisfaction reward. We additionally experiment with reward-weighted behavior cloning to further improve the constraint satisfaction rate of LMs. To evaluate our approach, we present a new Fine-grained Constraint Satisfaction (FineCS) benchmark, featuring two challenging tasks: (1) Text Style Transfer, where the goal is to simultaneously modify the sentiment and complexity of reviews, and (2) Protein Design, focusing on modulating fluorescence and stability of Green Fluorescent Proteins (GFP). Our empirical results show that MACS achieves the highest threshold satisfaction in both FineCS tasks, outperforming strong domain-specific baselines. Our work opens new avenues for generalized and real-value multi-attribute control, with implications for diverse applications spanning NLP and bioinformatics.

arXiv:2412.18697

Agents on the Bench: Large Language Model Based Multi Agent Framework for Trustworthy Digital Justice

The justice system has increasingly employed AI techniques to enhance efficiency, yet limitations remain in improving the quality of decision-making, particularly regarding transparency and explainability needed to uphold public trust in legal AI. To address these challenges, we propose a large language model based multi-agent framework named AgentsBench, which aims to simultaneously improve both efficiency and quality in judicial decision-making. Our approach leverages multiple LLM-driven agents that simulate the collaborative deliberation and decision making process of a judicial bench. We conducted experiments on legal judgment prediction task, and the results show that our framework outperforms existing LLM based methods in terms of performance and decision quality. By incorporating these elements, our framework reflects real-world judicial processes more closely, enhancing accuracy, fairness, and society consideration. AgentsBench provides a more nuanced and realistic methods of trustworthy AI decision-making, with strong potential for application across various case types and legal scenarios.

arXiv:2412.18619공식 코드 · Papers with Code

Next Token Prediction Towards Multimodal Intelligence: A Comprehensive Survey

Building on the foundations of language modeling in natural language processing, Next Token Prediction (NTP) has evolved into a versatile training objective for machine learning tasks across various modalities, achieving considerable success. As Large Language Models (LLMs) have advanced to unify understanding and generation tasks within the textual modality, recent research has shown that tasks from different modalities can also be effectively encapsulated within the NTP framework, transforming the multimodal information into tokens and predict the next one given the context. This survey introduces a comprehensive taxonomy that unifies both understanding and generation within multimodal learning through the lens of NTP. The proposed taxonomy covers five key aspects: Multimodal tokenization, MMNTP model architectures, unified task representation, datasets \& evaluation, and open challenges. This new taxonomy aims to aid researchers in their exploration of multimodal intelligence. An associated GitHub repository collecting the latest papers and repos is available at https://github.com/LMM101/Awesome-Multimodal-Next-Token-Prediction

arXiv:2412.19638공식 코드 · Papers with Code

Xmodel-2 Technical Report

Xmodel-2 is a 1.2-billion-parameter large language model designed specifically for reasoning tasks. Its architecture enables different model scales to share a unified set of hyperparameters, allowing for extensive experimentation on smaller models and seamless transfer of optimal configurations to larger models. To maximize training efficiency and stability, Xmodel-2 employs the WSD learning rate scheduler from MiniCPM. Pretrained on 1.5 trillion tokens from diverse sources, Xmodel-2 achieves state-of-the-art performance in complex reasoning and agent-based tasks, while maintaining low training costs. These results highlight the potential of efficient model design and training strategies in advancing reasoning capabilities. Model checkpoints and code are publicly available on GitHub at https://github.com/XiaoduoAILab/Xmodel-2

arXiv:2412.19394공식 코드 · Papers with Code

An Engorgio Prompt Makes Large Language Model Babble on

Auto-regressive large language models (LLMs) have yielded impressive performance in many real-world tasks. However, the new paradigm of these LLMs also exposes novel threats. In this paper, we explore their vulnerability to inference cost attacks, where a malicious user crafts Engorgio prompts to intentionally increase the computation cost and latency of the inference process. We design Engorgio, a novel methodology, to efficiently generate adversarial Engorgio prompts to affect the target LLM's service availability. Engorgio has the following two technical contributions. (1) We employ a parameterized distribution to track LLMs' prediction trajectory. (2) Targeting the auto-regressive nature of LLMs' inference process, we propose novel loss functions to stably suppress the appearance of the <EOS> token, whose occurrence will interrupt the LLM's generation process. We conduct extensive experiments on 13 open-sourced LLMs with parameters ranging from 125M to 30B. The results show that Engorgio prompts can successfully induce LLMs to generate abnormally long outputs (i.e., roughly 2-13$\times$ longer to reach 90%+ of the output length limit) in a white-box scenario and our real-world experiment demonstrates Engergio's threat to LLM service with limited computing resources. The code is released at: https://github.com/jianshuod/Engorgio-prompt.

arXiv:2412.19140공식 코드 · Papers with Code

SILC-EFSA: Self-aware In-context Learning Correction for Entity-level Financial Sentiment Analysis

In recent years, fine-grained sentiment analysis in finance has gained significant attention, but the scarcity of entity-level datasets remains a key challenge. To address this, we have constructed the largest English and Chinese financial entity-level sentiment analysis datasets to date. Building on this foundation, we propose a novel two-stage sentiment analysis approach called Self-aware In-context Learning Correction (SILC). The first stage involves fine-tuning a base large language model to generate pseudo-labeled data specific to our task. In the second stage, we train a correction model using a GNN-based example retriever, which is informed by the pseudo-labeled data. This two-stage strategy has allowed us to achieve state-of-the-art performance on the newly constructed datasets, advancing the field of financial sentiment analysis. In a case study, we demonstrate the enhanced practical utility of our data and methods in monitoring the cryptocurrency market. Our datasets and code are available at https://github.com/NLP-Bin/SILC-EFSA.

arXiv:2412.18541

PLD-Tree: Persistent Laplacian Decision Tree for Protein-Protein Binding Free Energy Prediction

Recent advances in topology-based modeling have accelerated progress in physical modeling and molecular studies, including applications to protein-ligand binding affinity. In this work, we introduce the Persistent Laplacian Decision Tree (PLD-Tree), a novel method designed to address the challenging task of predicting protein-protein interaction (PPI) affinities. PLD-Tree focuses on protein chains at binding interfaces and employs the persistent Laplacian to capture topological invariants reflecting critical inter-protein interactions. These topological descriptors, derived from persistent homology, are further enhanced by incorporating evolutionary scale modeling (ESM) from a large language model to integrate sequence-based information. We validate PLD-Tree on two benchmark datasets-PDBbind V2020 and SKEMPI v2 demonstrating a correlation coefficient ($R_p$) of 0.83 under the sophisticated leave-out-protein-out cross-validation. Notably, our approach outperforms all reported state-of-the-art methods on these datasets. These results underscore the power of integrating machine learning techniques with topology-based descriptors for molecular docking and virtual screening, providing a robust and accurate framework for predicting protein-protein binding affinities.

arXiv:2412.18275공식 코드 · Papers with Code

Learning to engineer protein flexibility

Generative machine learning models are increasingly being used to design novel proteins for therapeutic and biotechnological applications. However, the current methods mostly focus on the design of proteins with a fixed backbone structure, which leads to their limited ability to account for protein flexibility, one of the crucial properties for protein function. Learning to engineer protein flexibility is problematic because the available data are scarce, heterogeneous, and costly to obtain using computational as well as experimental methods. Our contributions to address this problem are three-fold. First, we comprehensively compare methods for quantifying protein flexibility and identify data relevant to learning. Second, we design and train flexibility predictors utilizing sequential or both sequential and structural information on the input. We overcome the data scarcity issue by leveraging a pre-trained protein language model. Third, we introduce a method for fine-tuning a protein inverse folding model to steer it toward desired flexibility in specified regions. We demonstrate that our method Flexpert-Design enables guidance of inverse folding models toward increased flexibility. This opens up new possibilities for protein flexibility engineering and the development of proteins with enhanced biological activities.

arXiv:2412.18603공식 코드 · Papers with Code

Long-Form Speech Generation with Spoken Language Models

We consider the generative modeling of speech over multiple minutes, a requirement for long-form multimedia generation and audio-native voice assistants. However, current spoken language models struggle to generate plausible speech past tens of seconds, from high temporal resolution of speech tokens causing loss of coherence, to architectural issues with long-sequence training or extrapolation, to memory costs at inference time. With these considerations we propose SpeechSSM, the first speech language model to learn from and sample long-form spoken audio (e.g., 16 minutes of read or extemporaneous speech) in a single decoding session without text intermediates, based on recent advances in linear-time sequence modeling. Furthermore, to address growing challenges in spoken language evaluation, especially in this new long-form setting, we propose: new embedding-based and LLM-judged metrics; quality measurements over length and time; and a new benchmark for long-form speech processing and generation, LibriSpeech-Long. Speech samples and the dataset are released at https://google.github.io/tacotron/publications/speechssm/

arXiv:2412.18566

Zero-resource Speech Translation and Recognition with LLMs

Despite recent advancements in speech processing, zero-resource speech translation (ST) and automatic speech recognition (ASR) remain challenging problems. In this work, we propose to leverage a multilingual Large Language Model (LLM) to perform ST and ASR in languages for which the model has never seen paired audio-text data. We achieve this by using a pre-trained multilingual speech encoder, a multilingual LLM, and a lightweight adaptation module that maps the audio representations to the token embedding space of the LLM. We perform several experiments both in ST and ASR to understand how to best train the model and what data has the most impact on performance in previously unseen languages. In ST, our best model is capable to achieve BLEU scores over 23 in CoVoST2 for two previously unseen languages, while in ASR, we achieve WERs of up to 28.2\%. We finally show that the performance of our system is bounded by the ability of the LLM to output text in the desired language.

arXiv:2412.18496공식 코드 · Papers with Code

Generating event descriptions under syntactic and semantic constraints

With the goal of supporting scalable lexical semantic annotation, analysis, and theorizing, we conduct a comprehensive evaluation of different methods for generating event descriptions under both syntactic constraints -- e.g. desired clause structure -- and semantic constraints -- e.g. desired verb sense. We compare three different methods -- (i) manual generation by experts; (ii) sampling from a corpus annotated for syntactic and semantic information; and (iii) sampling from a language model (LM) conditioned on syntactic and semantic information -- along three dimensions of the generated event descriptions: (a) naturalness, (b) typicality, and (c) distinctiveness. We find that all methods reliably produce natural, typical, and distinctive event descriptions, but that manual generation continues to produce event descriptions that are more natural, typical, and distinctive than the automated generation methods. We conclude that the automated methods we consider produce event descriptions of sufficient quality for use in downstream annotation and analysis insofar as the methods used for this annotation and analysis are robust to a small amount of degradation in the resulting event descriptions.

arXiv:2412.18443

Is Large Language Model Good at Triple Set Prediction? An Empirical Study

The core of the Knowledge Graph Completion (KGC) task is to predict and complete the missing relations or nodes in a KG. Common KGC tasks are mostly about inferring unknown elements with one or two elements being known in a triple. In comparison, the Triple Set Prediction (TSP) task is a more realistic knowledge graph completion task. It aims to predict all elements of unknown triples based on the information from known triples. In recent years, large language models (LLMs) have exhibited significant advancements in language comprehension, demonstrating considerable potential for KGC tasks. However, the potential of LLM on the TSP task has not yet to be investigated. Thus in this paper we proposed a new framework to explore the strengths and limitations of LLM in the TSP task. Specifically, the framework consists of LLM-based rule mining and LLM-based triple set prediction. The relation list of KG embedded within rich semantic information is first leveraged to prompt LLM in the generation of rules. This process is both efficient and independent of statistical information, making it easier to mine effective and realistic rules. For each subgraph, the specified rule is applied in conjunction with the relevant triples within that subgraph to guide the LLM in predicting the missing triples. Subsequently, the predictions from all subgraphs are consolidated to derive the complete set of predicted triples on KG. Finally, the method is evaluated on the relatively complete CFamily dataset. The experimental results indicate that when LLMs are required to adhere to a large amount of factual knowledge to predict missing triples, significant hallucinations occurs, leading to a noticeable decline in performance. To further explore the causes of this phenomenon, this paper presents a comprehensive analysis supported by a detailed case study.

arXiv:2412.18232

Efficient Long Context Language Model Retrieval with Compression

Long Context Language Models (LCLMs) have emerged as a new paradigm to perform Information Retrieval (IR), which enables the direct ingestion and retrieval of information by processing an entire corpus in their single context, showcasing the potential to surpass traditional sparse and dense retrieval methods. However, processing a large number of passages within in-context for retrieval is computationally expensive, and handling their representations during inference further exacerbates the processing time; thus, we aim to make LCLM retrieval more efficient and potentially more effective with passage compression. Specifically, we propose a new compression approach tailored for LCLM retrieval, which is trained to maximize the retrieval performance while minimizing the length of the compressed passages. To accomplish this, we generate the synthetic data, where compressed passages are automatically created and labeled as chosen or rejected according to their retrieval success for a given query, and we train the proposed Compression model for Long context Retrieval (CoLoR) with this data via preference optimization while adding the length regularization loss on top of it to enforce brevity. Through extensive experiments on 9 datasets, we show that CoLoR improves the retrieval performance by 6% while compressing the in-context size by a factor of 1.91.

arXiv:2412.18151

CoAM: Corpus of All-Type Multiword Expressions

Multiword expressions (MWEs) refer to idiomatic sequences of multiple words. MWE identification, i.e., detecting MWEs in text, can play a key role in downstream tasks such as machine translation. Existing datasets for MWE identification are inconsistently annotated, limited to a single type of MWE, or limited in size. To enable reliable and comprehensive evaluation, we created CoAM: Corpus of All-Type Multiword Expressions, a dataset of 1.3K sentences constructed through a multi-step process to enhance data quality consisting of human annotation, human review, and automated consistency checking. MWEs in CoAM are tagged with MWE types, such as Noun and Verb, to enable fine-grained error analysis. Annotations for CoAM were collected using a new interface created with our interface generator, which allows easy and flexible annotation of MWEs in any form, including discontinuous ones. Through experiments using CoAM, we find that a fine-tuned large language model outperforms the current state-of-the-art approach for MWE identification. Furthermore, analysis using our MWE type tagged data reveals that Verb MWEs are easier than Noun MWEs to identify across approaches.

arXiv:2412.18135

LSAQ: Layer-Specific Adaptive Quantization for Large Language Model Deployment

As Large Language Models (LLMs) demonstrate exceptional performance across various domains, deploying LLMs on edge devices has emerged as a new trend. Quantization techniques, which reduce the size and memory requirements of LLMs, are effective for deploying LLMs on resource-limited edge devices. However, existing one-size-fits-all quantization methods often fail to dynamically adjust the memory requirements of LLMs, limiting their applications to practical edge devices with various computation resources. To tackle this issue, we propose Layer-Specific Adaptive Quantization (LSAQ), a system for adaptive quantization and dynamic deployment of LLMs based on layer importance. Specifically, LSAQ evaluates the importance of LLMs' neural layers by constructing top-k token sets from the inputs and outputs of each layer and calculating their Jaccard similarity. Based on layer importance, our system adaptively adjusts quantization strategies in real time according to the computation resource of edge devices, which applies higher quantization precision to layers with higher importance, and vice versa. {Experimental results show that LSAQ consistently outperforms the selected quantization baselines in terms of perplexity and zero-shot tasks. Additionally, it can devise appropriate quantization schemes for different usage scenarios to facilitate the deployment of LLMs.

arXiv:2412.18093

Molly: Making Large Language Model Agents Solve Python Problem More Logically

Applying large language models (LLMs) as teaching assists has attracted much attention as an integral part of intelligent education, particularly in computing courses. To reduce the gap between the LLMs and the computer programming education expert, fine-tuning and retrieval augmented generation (RAG) are the two mainstream methods in existing researches. However, fine-tuning for specific tasks is resource-intensive and may diminish the model`s generalization capabilities. RAG can perform well on reducing the illusion of LLMs, but the generation of irrelevant factual content during reasoning can cause significant confusion for learners. To address these problems, we introduce the Molly agent, focusing on solving the proposed problem encountered by learners when learning Python programming language. Our agent automatically parse the learners' questioning intent through a scenario-based interaction, enabling precise retrieval of relevant documents from the constructed knowledge base. At generation stage, the agent reflect on the generated responses to ensure that they not only align with factual content but also effectively answer the user's queries. Extensive experimentation on a constructed Chinese Python QA dataset shows the effectiveness of the Molly agent, indicating an enhancement in its performance for providing useful responses to Python questions.

arXiv:2412.17921

VITRO: Vocabulary Inversion for Time-series Representation Optimization

Although LLMs have demonstrated remarkable capabilities in processing and generating textual data, their pre-trained vocabularies are ill-suited for capturing the nuanced temporal dynamics and patterns inherent in time series. The discrete, symbolic nature of natural language tokens, which these vocabularies are designed to represent, does not align well with the continuous, numerical nature of time series data. To address this fundamental limitation, we propose VITRO. Our method adapts textual inversion optimization from the vision-language domain in order to learn a new time series per-dataset vocabulary that bridges the gap between the discrete, semantic nature of natural language and the continuous, numerical nature of time series data. We show that learnable time series-specific pseudo-word embeddings represent time series data better than existing general language model vocabularies, with VITRO-enhanced methods achieving state-of-the-art performance in long-term forecasting across most datasets.