T/ZGTXXH 076-2023 The workload requirements for assessing the entire machine and cluster computing power of artificial intelligence training servers

Scope

Main Technical Content: 1 Scope 2 Normative Reference Documents 3 Terms and Definitions 4 Abbreviations 5 Workload Indicators 5.1 Training Scenario Classification 5.1.1 Image Classification 5.1.2 Object Detection 5.1.3 Semantic Segmentation 5.1.4 Instance Segmentation 5.1.5 Speech Recognition 5.1.6 Text Recognition 5.1.7 Machine Translation 5.1.8 Recommendation 5.1.9 Reinforcement Learning 5.1.10 AIGC (Artificial Intelligence Generative Content) 5.2 Model Evaluation Indicators 5.2.1 Completeness 5.2.2 Accuracy 5.2.3 Compatibility 5.2.4 Portability 5.2.5 Interpretability 5.2.6 Robustness 5.2.7 Efficiency 5.2.8 Model Scale 5.2.9 Reproducibility 5.2.10 Scalability 5.2.11 Parallelism 5.3 Data Set Evaluation Indicators 5.3.1 Data Set Scale 5.3.2 Data Diversity 5.3.3 Data Balance 5.3.4 Data Set Annotation Quality 5.3.5 Data Set Contamination 5.3.6 Data Quality 5.3.7 Data Timeliness (Update) 6 Requirements for the Workload of AI Training Servers 6.1 Model Requirements 6.1.1 Functional Requirements 6.1.2 Non-functional Requirements 6.2 Data Set Requirements 6.2.1 Scale Requirements 6.2.2 Diversity Requirements 6.2.3 Balance Requirements 6.2.4 Annotation Quality Requirements 6.2.5 Contamination Requirements 6.2.6 Data Quality Requirements 6.2.7 Timeliness Requirements Appendix A (Informative) Examples of Workload Applications for Evaluating the Computing Power of AI Training Servers and Clusters A.1 Workload Application Examples A.2 Workload Application Requirements A.3 Data Set Requirements A.4 Model Requirements A.5 Data Set Evaluation Indicators A.6 Data Set Quality Requirements A.7 Data Set Timeliness Requirements A.8 Data Set Pollution Requirements A.9 Data Set Quality Requirements A.10 Data Set Efficiency Requirements A.11 Parallelis

Development Information

Related Standards