DC Field | Value | Language |
---|---|---|
dc.contributor.advisor | 김주영 | - |
dc.contributor.author | Lee, Sukjin | - |
dc.contributor.author | 이석진 | - |
dc.date.accessioned | 2024-07-30T19:31:24Z | - |
dc.date.available | 2024-07-30T19:31:24Z | - |
dc.date.issued | 2024 | - |
dc.identifier.uri | http://library.kaist.ac.kr/search/detail/view.do?bibCtrlNo=1096796&flag=dissertation | en_US |
dc.identifier.uri | http://hdl.handle.net/10203/321578 | - |
dc.description | 학위논문(석사) - 한국과학기술원 : 전기및전자공학부, 2024.2,[ii, 21 p. :] | - |
dc.description.abstract | This paper presents SparsePU, a processing unit capable of leveraging bit-slice level sparsity for accelerating transformer models within a processing-in-memory (PIM) architecture. This processor achieves performance enhancements by utilizing both activation and weight unstructured bit-slice level sparsity, which has been challenging in conventional PIM structures. The proposed accelerator accelerates operations by performing row-wise matrix multiplication for activation sparsity and enables simultaneous acceleration for various ratios of weight sparsity through a row-wise compressed weight data format. It integrates a network within the accelerator for effective and efficient accumulation of compressed weight data. Additionally, it maximizes operational acceleration in high activation sparsity scenarios through a multi-row skipping scheme. The accelerator significantly enhances performance, achieving up to 857.27x faster computation in actual transformer model layers, and reduces the size of sparse weight data to be stored by up to 93.68%. | - |
dc.language | eng | - |
dc.publisher | 한국과학기술원 | - |
dc.subject | 프로세싱-인-메모리▼a트랜스포머 모델▼a비트-슬라이스 레벨 희소성 | - |
dc.subject | Processing-in-memory(PIM)▼aTransformer model▼aBit-slice level sparsity | - |
dc.title | (A) PIM-based SparsePU for transformer model acceleration with bit-slice level sparsity exploitation | - |
dc.title.alternative | 비트 슬라이스 레벨 희소성 활용을 통한 트랜스포머 모델 가속 프로세싱-인-메모리 기반 희소 연산 유닛 | - |
dc.type | Thesis(Master) | - |
dc.identifier.CNRN | 325007 | - |
dc.description.department | 한국과학기술원 :전기및전자공학부, | - |
dc.contributor.alternativeauthor | Kim, Joo-Young | - |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.