Image de couverture de livre

UniAttn: Reducing Inference Costs via Softmax Unification for Post-Training LLMs

Fuente: arXiv

Enregistré dans:

Détails bibliographiques
Auteurs principaux:	Xiong, Yizhe, Huang, Wei, Ye, Xin, Chen, Hui, Lin, Zijia, Lian, Haoran, Su, Zhenpeng, Han, Jungong, Ding, Guiguang
Format:	Preprint
Publié:	2025
Sujets:	Computation and Language
Accès en ligne:	Acceder al recurso
Tags:	Ajouter un tag Pas de tags, Soyez le premier à ajouter un tag!

Documents similaires

Temporal Scaling Law for Large Language Models
par: Xiong, Yizhe, et autres
Publié: (2024)

Scaffold-BPE: Enhancing Byte Pair Encoding for Large Language Models with Simple and Effective Scaffold Token Removal
par: Lian, Haoran, et autres
Publié: (2024)

PYRA: Parallel Yielding Re-Activation for Training-Inference Efficient Task Adaptation
par: Xiong, Yizhe, et autres
Publié: (2024)

Finedeep: Mitigating Sparse Activation in Dense LLMs via Multi-Layer Fine-Grained Experts
par: Pan, Leiyu, et autres
Publié: (2025)

DSMoE: Matrix-Partitioned Experts with Dynamic Routing for Computation-Efficient Dense LLMs
par: Lv, Minxuan, et autres
Publié: (2025)

CAIT: Triple-Win Compression towards High Accuracy, Fast Inference, and Favorable Transferability For ViTs
par: Wang, Ao, et autres
Publié: (2023)

[CLS] Token Tells Everything Needed for Training-free Efficient MLLMs
par: Wang, Ao, et autres
Publié: (2024)

RepViT-SAM: Towards Real-Time Segmenting Anything
par: Wang, Ao, et autres
Publié: (2023)

LSNet: See Large, Focus Small
par: Wang, Ao, et autres
Publié: (2025)

RepViT: Revisiting Mobile CNN From ViT Perspective
par: Wang, Ao, et autres
Publié: (2023)

MaskMoE: Boosting Token-Level Learning via Routing Mask in Mixture-of-Experts
par: Su, Zhenpeng, et autres
Publié: (2024)

LBPE: Long-token-first Tokenization to Improve Large Language Models
par: Lian, Haoran, et autres
Publié: (2024)

Neutralizing Token Aggregation via Information Augmentation for Efficient Test-Time Adaptation
par: Xiong, Yizhe, et autres
Publié: (2025)

Fast Quiet-STaR: Thinking Without Thought Tokens
par: Huang, Wei, et autres
Publié: (2025)

CartesianMoE: Boosting Knowledge Sharing among Experts via Cartesian Product Routing in Mixture-of-Experts
par: Su, Zhenpeng, et autres
Publié: (2024)

YOLOE: Real-Time Seeing Anything
par: Wang, Ao, et autres
Publié: (2025)

Advancing Reliable Test-Time Adaptation of Vision-Language Models under Visual Variations
par: Liang, Yiwen, et autres
Publié: (2025)

YOLO-UniOW: Efficient Universal Open-World Object Detection
par: Liu, Lihao, et autres
Publié: (2024)

Learn from the Learnt: Source-Free Active Domain Adaptation via Contrastive Sampling and Visual Persistence
par: Lyu, Mengyao, et autres
Publié: (2024)

YOLOv10: Real-Time End-to-End Object Detection
par: Wang, Ao, et autres
Publié: (2024)

Context Enhancement with Reconstruction as Sequence for Unified Unsupervised Anomaly Detection
par: Yang, Hui-Yue, et autres
Publié: (2024)

Breaking the Stage Barrier: A Novel Single-Stage Approach to Long Context Extension for Large Language Models
par: Lian, Haoran, et autres
Publié: (2024)

MiLe Loss: a New Entropy-Weighed Loss for Mitigating the Bias of Learning Difficulties in Large Language Models
par: Su, Zhenpeng, et autres
Publié: (2023)

Tracking and Segmenting Anything in Any Modality
par: Zhang, Tianlu, et autres
Publié: (2025)

PrefixKV: Adaptive Prefix KV Cache is What Vision Instruction-Following Models Need for Efficient Generation
par: Wang, Ao, et autres
Publié: (2024)

PruneHal: Reducing Hallucinations in Multi-modal Large Language Models through Adaptive KV Cache Pruning
par: Sun, Fengyuan, et autres
Publié: (2025)

AdaTP: Attention-Debiased Token Pruning for Video Large Language Models
par: Sun, Fengyuan, et autres
Publié: (2025)

Promptable Anomaly Segmentation with SAM Through Self-Perception Tuning
par: Yang, Hui-Yue, et autres
Publié: (2024)

Towards Efficient Vision-Language Tuning: More Information Density, More Generalizability
par: Hao, Tianxiang, et autres
Publié: (2023)

LLMI3D: MLLM-based 3D Perception from a Single 2D Image
par: Yang, Fan, et autres
Publié: (2024)

Cream of the Crop: Harvesting Rich, Scalable and Transferable Multi-Modal Data for Instruction Fine-Tuning
par: Lyu, Mengyao, et autres
Publié: (2025)

One-Dimensional Adapter to Rule Them All: Concepts, Diffusion Models and Erasing Applications
par: Lyu, Mengyao, et autres
Publié: (2023)

AttnRouter: Per-Category Attention Routing for Training-Free Image Editing on MMDiT
par: Li, Guandong, et autres
Publié: (2026)

Self-Adjust Softmax
par: Zheng, Chuanyang, et autres
Publié: (2025)

Is Softmax Loss All You Need? A Principled Analysis of Softmax-family Loss
par: Pu, Yuanhao, et autres
Publié: (2026)

AttnMod: Attention-Based New Art Styles
par: Su, Shih-Chieh
Publié: (2024)

Attn-GS: Attention-Guided Context Compression for Efficient Personalized LLMs
par: Zeng, Shenglai, et autres
Publié: (2026)

End-to-End On-Device Quantization-Aware Training for LLMs at Inference Cost
par: Tan, Qitao, et autres
Publié: (2025)

AttnCache: Accelerating Self-Attention Inference for LLM Prefill via Attention Cache
par: Song, Dinghong, et autres
Publié: (2025)

AttnGCG: Enhancing Jailbreaking Attacks on LLMs with Attention Manipulation
par: Wang, Zijun, et autres
Publié: (2024)