A Survey of Metrics to Enhance Training Dependability in Large Language Models
- Resource Type
- Conference
- Authors
- Fang, Wenyi; Zhang, Hao; Gong, Ziyu; Zeng, Longbin; Lu, Xuhui; Liu, Biao; Wu, Xiaoyu; Zheng, Yang; Hu, Zheng; Zhang, Xun
- Source
- 2023 IEEE 34th International Symposium on Software Reliability Engineering Workshops (ISSREW) ISSREW Software Reliability Engineering Workshops (ISSREW), 2023 IEEE 34th International Symposium on. :180-185 Oct, 2023
- Subject
- Aerospace
Communication, Networking and Broadcast Technologies
Computing and Processing
Nuclear Engineering
Power, Energy and Industry Applications
Robotics and Control Systems
Transportation
Training
Measurement
Surveys
Systematics
Focusing
Stability analysis
Software reliability
Large Language Model
Dependability
Monitoring Metric
- Language
The rapidly advancing field of artificial intelligence requires meticulous attention to the training and monitoring of large language models (LLMs). This paper offers a systematic analysis of existing metrics and introduces new ones, focusing on their theoretical underpinnings and practical implementations. We present empirical results and insights into the performance of selected metrics, elucidating the complex interplay of variables in the training process. Our comprehensive approach provides significant insights into LLM training, and promises to improve the dependability and efficiency of future models.