Operating Conversational Large Language Models (LLMs)in the Presenceof Errors

被引:0
|
作者
Gao, Zhen [1 ]
Deng, Jie [2 ]
Reviriego, Pedro [3 ]
Liu, Shanshan [4 ]
Pozo, Alejando [3 ]
Lombardi, Fabrizio [5 ]
机构
[1] Tianjin Univ, Sch Elect & Informat Engn, Tianjin 300072, Peoples R China
[2] Tianjin Univ, Sch Future Technol, Tianjin 300072, Peoples R China
[3] Univ Politecn Madrid, ETSI Telecomunicac, Madrid 28040, Spain
[4] Univ Elect Sci & Technol China, Sch Informat & Commun Engn, Chengdu 611731, Peoples R China
[5] Northeastern Univ, Dept Elect & Comp Engn, Boston, MA 02215 USA
关键词
Quantization (signal); Benchmark testing; Transformers; Codes; Translation; Memory management; Logic gates; Integrated circuit modeling; Hardware; Computational modeling; Dependability; generative artificial intelligence; large language models; errors;
D O I
10.1109/MNANO.2024.3513112
中图分类号
TB3 [工程材料学];
学科分类号
0805 ; 080502 ;
摘要
Conversational Large Language Models have taken the center stage of the artificial intelligence landscape. As they are pervasive, there is a need to evaluate their dependability, i.e., performance when errors appear due to the underlying hardware implementation. In this paper we consider the evaluation of the dependability of a widely used conversational LLM: Mistral-7B. Error injection is conducted, and the Multitask Language Understanding (MMLU) benchmark is used to evaluate the impact on performance. The drop in the percentage of correct answers due to errors is analyzed and the results provide interesting insights: Mistral-7B has a large intrinsic tolerance to errors even at high bit error rates. This opens the door to the use of nanotechnologies that trade-off errors for energy dissipation and complexity to further improve the LLM implementation. Also, the error tolerance is larger for 8-bit quantization than for 4-bit quantization, so suggesting that there will be also a trade-off between quantization optimizations to reduce memory requirements and error tolerance. In addition, we also show the different impact of errors on different types of weights, which is valuable information for selective protection designs.
引用
收藏
页码:31 / 37
页数:7
相关论文
共 50 条
  • [21] Large Language Models (LLMs) Enable Few-Shot Clustering
    Vijay, Viswanathan
    Kiril, Gashteovski
    Carolin, Lawrence
    Tongshuang, Wu
    Graham, Neubig
    NEC Technical Journal, 2024, 17 (02): : 80 - 90
  • [22] LLMs to the Moon? Reddit Market Sentiment Analysis with Large Language Models
    Deng, Xiang
    Bashlovkina, Vasilisa
    Han, Feng
    Baumgartner, Simon
    Bendersky, Michael
    COMPANION OF THE WORLD WIDE WEB CONFERENCE, WWW 2023, 2023, : 1014 - 1019
  • [23] Reducing the Energy Dissipation of Large Language Models (LLMs) with Approximate Memories
    Gao, Zhen
    Deng, Jie
    Reviriego, Pedro
    Liu, Shanshan
    Lombardi, Fabrizio
    2024 IEEE INTERNATIONAL SYMPOSIUM ON CIRCUITS AND SYSTEMS, ISCAS 2024, 2024,
  • [24] Leveraging Large Language Models (LLMs) For Randomized Clinical Trial Summarization
    Mangla, Anjali
    Thangaraj, Phyllis
    Khera, Rohan
    CIRCULATION, 2024, 150
  • [25] Reinforcement Learning With Large Language Models (LLMs) Interaction For Network Services
    Du, Hongyang
    Zhang, Ruichen
    Niyato, Dusit
    Kang, Jiawen
    Xiong, Zehui
    Kim, Dong In
    2024 INTERNATIONAL CONFERENCE ON COMPUTING, NETWORKING AND COMMUNICATIONS, ICNC, 2024, : 799 - 803
  • [26] Towards trustworthy LLMs: a review on debiasing and dehallucinating in large language models
    Lin, Zichao
    Guan, Shuyan
    Zhang, Wending
    Zhang, Huiyan
    Li, Yugang
    Zhang, Huaping
    ARTIFICIAL INTELLIGENCE REVIEW, 2024, 57 (09)
  • [27] A Review of Current Trends, Techniques, and Challenges in Large Language Models (LLMs)
    Patil, Rajvardhan
    Gudivada, Venkat
    APPLIED SCIENCES-BASEL, 2024, 14 (05):
  • [28] Performance of large language models (LLMs) in providing prostate cancer information
    Alasker, Ahmed
    Alsalamah, Seham
    Alshathri, Nada
    Almansour, Nura
    Alsalamah, Faris
    Alghafees, Mohammad
    Alkhamees, Mohammad
    Alsaikhan, Bader
    BMC UROLOGY, 2024, 24 (01):
  • [29] Enhancing Accessibility in Software Engineering Projects with Large Language Models (LLMs)
    Aljedaani, Wajdi
    Eler, Marcelo Medeiros
    Parthasarathy, P. D.
    PROCEEDINGS OF THE 56TH ACM TECHNICAL SYMPOSIUM ON COMPUTER SCIENCE EDUCATION, SIGCSE TS 2025, VOL 1, 2025, : 25 - 31
  • [30] Large language models (LLMs): survey, technical frameworks, and future challenges
    Kumar, Pranjal
    ARTIFICIAL INTELLIGENCE REVIEW, 2024, 57 (09)