Enhancing Large Language Models with Structured Data: A Unified Hypergraph Approach (LLaSA)
LLaSA is a novel framework that improves Large Language Models' ability to process and utilize structured data (tables, graphs, databases) by representing them as hypergraphs, enabling a unified encoding method and enhancing performance on various knowledge-intensive tasks.
WavChat：口語對話模型綜述
本文全面概述了口語對話模型，特別是級聯和端到端模型，並深入探討了語音表徵、訓練範式、串流、雙工和互動能力等核心技術，以及相關數據集、評估指標和基準。
Introducing WORLDREP: A Large Language Model-Powered Dataset for Predicting Future International Events from Text
WORLDREP is a new dataset that leverages the power of large language models to predict future international events from text, addressing limitations of existing datasets by capturing complex multilateral relations and providing high-quality, expert-validated labels.
AddrLLM: An Innovative Large Language Model Framework for Rewriting Addresses Using Nationwide Logistics Data
AddrLLM, a novel framework leveraging retrieval-augmented large language models, effectively rewrites inaccurate addresses, significantly improving logistics efficiency.
METEOR: A Method for Evolving Large Language Models into Domain Experts
Large language models (LLMs) can be effectively specialized for specific domains using a multi-stage training approach that involves knowledge distillation, iterative refinement with expert feedback, and self-evolution through inference strategy optimization.
향상된 코드 생성을 위한 몬테카를로 트리 탐색 기반 자기 주도적 추론 증강: SRA-MCTS
대규모 언어 모델(LLM)의 코드 생성 능력을 향상시키기 위해 몬테카를로 트리 탐색(MCTS) 기반 자기 주도적 추론 증강 방식인 SRA-MCTS를 제안하며, 이는 다양한 중간 추론 경로를 생성하여 모델의 자율적인 사고를 촉진하고, 특히 소규모 모델의 성능을 크게 향상시킵니다.
Enhancing Code Generation in Large Language Models Through Self-Driven Reasoning Augmentation with Monte Carlo Tree Search (SRA-MCTS)
Integrating a self-driven reasoning augmentation process using Monte Carlo Tree Search (SRA-MCTS) significantly improves the code generation capabilities of large language models, particularly in solving complex problems, by enabling the models to autonomously generate and evaluate diverse reasoning paths.
BlueLM-V-3B: Optimizing Multimodal Large Language Models for Mobile Devices Through Algorithm and System Co-Design
BlueLM-V-3B is an efficient algorithm and system co-design approach that enables the deployment of powerful and fast multimodal large language models (MLLMs) on mobile devices by addressing the challenges of limited memory and computational resources.
AmoebaLLM: A Framework for Efficiently Deploying Large Language Models with Adaptable Structures
AmoebaLLM is a novel framework that enables the efficient deployment of large language models (LLMs) by allowing for the instant derivation of compressed subnets with arbitrary shapes, achieving a balance between accuracy and efficiency without the need for individual fine-tuning.
말뭉치 필터링을 통한 암시적 특징 정렬을 활용한 관점 기반 감성 분석: 저자원 환경에서의 한국어 리뷰 분석을 위한 KPC-cF 프레임워크 소개
본 논문은 한국어와 같이 자원이 부족한 언어에서 효과적인 관점 기반 감성 분석(ABSA)을 수행하기 위해 번역된 벤치마크 데이터와 레이블링 되지 않은 데이터를 활용하는 KPC-cF 프레임워크를 제안합니다.
© 2024 by Linnk AI