Cover Image

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

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Xiong, Yizhe, Huang, Wei, Ye, Xin, Chen, Hui, Lin, Zijia, Lian, Haoran, Su, Zhenpeng, Han, Jungong, Ding, Guiguang
Format:	Preprint
Published:	2025
Subjects:	Computation and Language
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

Temporal Scaling Law for Large Language Models
by: Xiong, Yizhe, et al.
Published: (2024)

Scaffold-BPE: Enhancing Byte Pair Encoding for Large Language Models with Simple and Effective Scaffold Token Removal
by: Lian, Haoran, et al.
Published: (2024)

PYRA: Parallel Yielding Re-Activation for Training-Inference Efficient Task Adaptation
by: Xiong, Yizhe, et al.
Published: (2024)

Finedeep: Mitigating Sparse Activation in Dense LLMs via Multi-Layer Fine-Grained Experts
by: Pan, Leiyu, et al.
Published: (2025)

DSMoE: Matrix-Partitioned Experts with Dynamic Routing for Computation-Efficient Dense LLMs
by: Lv, Minxuan, et al.
Published: (2025)

CAIT: Triple-Win Compression towards High Accuracy, Fast Inference, and Favorable Transferability For ViTs
by: Wang, Ao, et al.
Published: (2023)

[CLS] Token Tells Everything Needed for Training-free Efficient MLLMs
by: Wang, Ao, et al.
Published: (2024)

RepViT-SAM: Towards Real-Time Segmenting Anything
by: Wang, Ao, et al.
Published: (2023)

LSNet: See Large, Focus Small
by: Wang, Ao, et al.
Published: (2025)

RepViT: Revisiting Mobile CNN From ViT Perspective
by: Wang, Ao, et al.
Published: (2023)

MaskMoE: Boosting Token-Level Learning via Routing Mask in Mixture-of-Experts
by: Su, Zhenpeng, et al.
Published: (2024)

LBPE: Long-token-first Tokenization to Improve Large Language Models
by: Lian, Haoran, et al.
Published: (2024)

Neutralizing Token Aggregation via Information Augmentation for Efficient Test-Time Adaptation
by: Xiong, Yizhe, et al.
Published: (2025)

Fast Quiet-STaR: Thinking Without Thought Tokens
by: Huang, Wei, et al.
Published: (2025)

CartesianMoE: Boosting Knowledge Sharing among Experts via Cartesian Product Routing in Mixture-of-Experts
by: Su, Zhenpeng, et al.
Published: (2024)

YOLOE: Real-Time Seeing Anything
by: Wang, Ao, et al.
Published: (2025)

Advancing Reliable Test-Time Adaptation of Vision-Language Models under Visual Variations
by: Liang, Yiwen, et al.
Published: (2025)

YOLO-UniOW: Efficient Universal Open-World Object Detection
by: Liu, Lihao, et al.
Published: (2024)

Learn from the Learnt: Source-Free Active Domain Adaptation via Contrastive Sampling and Visual Persistence
by: Lyu, Mengyao, et al.
Published: (2024)

YOLOv10: Real-Time End-to-End Object Detection
by: Wang, Ao, et al.
Published: (2024)

Context Enhancement with Reconstruction as Sequence for Unified Unsupervised Anomaly Detection
by: Yang, Hui-Yue, et al.
Published: (2024)

Breaking the Stage Barrier: A Novel Single-Stage Approach to Long Context Extension for Large Language Models
by: Lian, Haoran, et al.
Published: (2024)

MiLe Loss: a New Entropy-Weighed Loss for Mitigating the Bias of Learning Difficulties in Large Language Models
by: Su, Zhenpeng, et al.
Published: (2023)

Tracking and Segmenting Anything in Any Modality
by: Zhang, Tianlu, et al.
Published: (2025)

PrefixKV: Adaptive Prefix KV Cache is What Vision Instruction-Following Models Need for Efficient Generation
by: Wang, Ao, et al.
Published: (2024)

PruneHal: Reducing Hallucinations in Multi-modal Large Language Models through Adaptive KV Cache Pruning
by: Sun, Fengyuan, et al.
Published: (2025)

AdaTP: Attention-Debiased Token Pruning for Video Large Language Models
by: Sun, Fengyuan, et al.
Published: (2025)

Promptable Anomaly Segmentation with SAM Through Self-Perception Tuning
by: Yang, Hui-Yue, et al.
Published: (2024)

Towards Efficient Vision-Language Tuning: More Information Density, More Generalizability
by: Hao, Tianxiang, et al.
Published: (2023)

LLMI3D: MLLM-based 3D Perception from a Single 2D Image
by: Yang, Fan, et al.
Published: (2024)

Cream of the Crop: Harvesting Rich, Scalable and Transferable Multi-Modal Data for Instruction Fine-Tuning
by: Lyu, Mengyao, et al.
Published: (2025)

One-Dimensional Adapter to Rule Them All: Concepts, Diffusion Models and Erasing Applications
by: Lyu, Mengyao, et al.
Published: (2023)

AttnRouter: Per-Category Attention Routing for Training-Free Image Editing on MMDiT
by: Li, Guandong, et al.
Published: (2026)

Self-Adjust Softmax
by: Zheng, Chuanyang, et al.
Published: (2025)

Is Softmax Loss All You Need? A Principled Analysis of Softmax-family Loss
by: Pu, Yuanhao, et al.
Published: (2026)

AttnMod: Attention-Based New Art Styles
by: Su, Shih-Chieh
Published: (2024)

Attn-GS: Attention-Guided Context Compression for Efficient Personalized LLMs
by: Zeng, Shenglai, et al.
Published: (2026)

End-to-End On-Device Quantization-Aware Training for LLMs at Inference Cost
by: Tan, Qitao, et al.
Published: (2025)

AttnCache: Accelerating Self-Attention Inference for LLM Prefill via Attention Cache
by: Song, Dinghong, et al.
Published: (2025)

AttnGCG: Enhancing Jailbreaking Attacks on LLMs with Attention Manipulation
by: Wang, Zijun, et al.
Published: (2024)