7 papers
UnionSparse: An Index-Efficient Sparsity Framework for Low-Bit Sparse LLM Inference on Edge
Tianhao Jiang, Hang Gu, Teng Wang +9
Edge LLM inference combines sparsity and low-bit quantization to meet device memory, latency, and power limits. Yet quantization shrinks weight payloads without proportionally redu…
Realizable N:M Sparse Transformer Inference via Search-Kernel Co-Design
Yiming Liu, Wenqi Lou, Zhiguang Wang +4
The paper presents a co-designed hardware and software framework that enables fast inference of Vision Transformers by applying N:M structured sparsity with a specialized CUDA kern…
A Timing-Anomaly Free Dynamic Scheduling on Heterogeneous Systems
Yixuan Zhu, Yinkang Gao, Lei Gong +9
Heterogeneous systems commonly adopt dynamic scheduling algorithms to improve resource utilization and enhance scheduling flexibility. However, such flexibility may introduce timin…
Reducing End-to-End Latency of Cause-Effect Chains with Shared Cache Analysis
Yixuan Zhu, Yinkang Gao, Bo Zhang +8
Cause-effect chains, as a widely used modeling method in real-time embedded systems, are extensively applied in various safety-critical domains. End-to-end latency, as a key real-t…
Crystal-KV: Efficient KV Cache Management for Chain-of-Thought LLMs via Answer-First Principle
Zihan Wang, Cheng Tang, Lei Gong +5
Chain-of-Thought (CoT) reasoning in large language models (LLMs) significantly improves accuracy on complex tasks, yet incurs excessive memory overhead due to the long think-stage…
CoQMoE: Co-Designed Quantization and Computation Orchestration for Mixture-of-Experts Vision Transformer on FPGA
Jiale Dong, Hao Wu, Zihao Wang +5
Vision Transformers (ViTs) exhibit superior performance in computer vision tasks but face deployment challenges on resource-constrained devices due to high computational/memory dem…