↓Skip to main content

2025-02-20s

2025

Why Safeguarded Ships Run Aground? Aligned Large Language Models' Safety Mechanisms Tend to Be Anchored in The Template Region
·2482 words·12 mins· loading · loading
AI Generated 🤗 Daily Papers AI Theory Safety 🏢 Hong Kong Polytechnic University
Aligned LLMs’ safety often anchors in the template region, creating vulnerabilities. Detaching safety mechanisms shows promise in mitigation.
Train Small, Infer Large: Memory-Efficient LoRA Training for Large Language Models
·3075 words·15 mins· loading · loading
AI Generated 🤗 Daily Papers Natural Language Processing Large Language Models 🏢 Zhejiang University
LORAM: Train small, infer large LLMs by memory-efficient LoRA training. Enables 70B parameter model training on a 20G HBM GPU, replacing A100-80G. Reduces parameter storage cost by 15.81x.
REFIND: Retrieval-Augmented Factuality Hallucination Detection in Large Language Models
·582 words·3 mins· loading · loading
AI Generated 🤗 Daily Papers Natural Language Processing Large Language Models 🏢 Pohang University of Science and Technology
REFIND: Detects LLM hallucinations by directly leveraging retrieved documents, using a novel Context Sensitivity Ratio.
Noise May Contain Transferable Knowledge: Understanding Semi-supervised Heterogeneous Domain Adaptation from an Empirical Perspective
·6916 words·33 mins· loading · loading
AI Generated 🤗 Daily Papers Machine Learning Transfer Learning 🏢 Beijing Teleinfo Technology Company Ltd., China Academy of Information and Communications Technology
Unveiling the surprising potential of noise: transferable knowledge in semi-supervised heterogeneous domain adaptation (SHDA).
MoM: Linear Sequence Modeling with Mixture-of-Memories
·2764 words·13 mins· loading · loading
AI Generated 🤗 Daily Papers Natural Language Processing Large Language Models 🏢 Shanghai AI Laboratory
MoM: Enhancing linear sequence modeling via mixture-of-memories for improved recall and reduced memory interference.
LongPO: Long Context Self-Evolution of Large Language Models through Short-to-Long Preference Optimization
·2370 words·12 mins· loading · loading
AI Generated 🤗 Daily Papers Natural Language Processing Large Language Models 🏢 National University of Singapore
LongPO: Self-evolve LLMs to excel in long contexts via short-to-long preference optimization, boosting performance without sacrificing short-context skills.
Is That Your Final Answer? Test-Time Scaling Improves Selective Question Answering
·2478 words·12 mins· loading · loading
AI Generated 🤗 Daily Papers Natural Language Processing Question Answering 🏢 Johns Hopkins University
Test-time scaling + confidence = better QA!
Craw4LLM: Efficient Web Crawling for LLM Pretraining
·3024 words·15 mins· loading · loading
AI Generated 🤗 Daily Papers Natural Language Processing Large Language Models 🏢 Tsinghua University
CRAW4LLM: Efficiently crawls web pages for LLM pretraining by prioritizing influence scores, boosting data quality & cutting crawling waste.
Autellix: An Efficient Serving Engine for LLM Agents as General Programs
·4705 words·23 mins· loading · loading
AI Generated 🤗 Daily Papers Natural Language Processing Large Language Models 🏢 UC Berkeley
Autellix: Efficient LLM Serving for Agents
AdaptiveStep: Automatically Dividing Reasoning Step through Model Confidence
·4758 words·23 mins· loading · loading
AI Generated 🤗 Daily Papers Machine Learning Reinforcement Learning 🏢 Nanjing University
AdaptiveStep: Divides reasoning steps automatically through model confidence, enhancing PRM training & performance.
SongGen: A Single Stage Auto-regressive Transformer for Text-to-Song Generation
·2399 words·12 mins· loading · loading
AI Generated 🤗 Daily Papers Speech and Audio Music Generation 🏢 Beihang University
SongGen: Single-stage autoregressive transformer for controllable text-to-song generation, simplifying the process and improving control.
RAD: Training an End-to-End Driving Policy via Large-Scale 3DGS-based Reinforcement Learning
·2823 words·14 mins· loading · loading
AI Generated 🤗 Daily Papers AI Applications Autonomous Vehicles 🏢 Huazhong University of Science & Technology
RAD: 3DGS-based RL advances autonomous driving, achieving a 3x lower collision rate!
NExT-Mol: 3D Diffusion Meets 1D Language Modeling for 3D Molecule Generation
·6586 words·31 mins· loading · loading
AI Generated 🤗 Daily Papers Machine Learning Deep Learning 🏢 National University of Singapore
NExT-Mol: Combines 1D language models with 3D diffusion for molecule generation, achieving state-of-the-art performance and validity.
Thinking Preference Optimization
·5794 words·28 mins· loading · loading
AI Generated 🤗 Daily Papers Machine Learning Deep Learning 🏢 Case.edu
ThinkPO improves LLM reasoning by preferring longer CoT, boosting performance without new data.
Small Models Struggle to Learn from Strong Reasoners
·4149 words·20 mins· loading · loading
AI Generated 🤗 Daily Papers Machine Learning Deep Learning 🏢 University of Washington
Small language models struggle to learn complex reasoning from large models, but a novel ‘Mix Distillation’ method balances complexity for effective capability transfer.
Presumed Cultural Identity: How Names Shape LLM Responses
·2724 words·13 mins· loading · loading
AI Generated 🤗 Daily Papers AI Theory Fairness 🏢 University of Copenhagen
LLMs personalize based on user names, but this study reveals that cultural presumptions in LLM responses risk reinforcing stereotypes.
InfiR : Crafting Effective Small Language Models and Multimodal Small Language Models in Reasoning
·1563 words·8 mins· loading · loading
AI Generated 🤗 Daily Papers Multimodal Learning Multimodal Reasoning 🏢 Reallm Labs
InfiR: Efficient, small AI models rival larger ones in reasoning, slashing costs and boosting privacy for wider AI use.