Skip to content
This repository has been archived by the owner on Jul 18, 2024. It is now read-only.

Latest commit

 

History

History
14 lines (11 loc) · 892 Bytes

sda_model_performance.md

File metadata and controls

14 lines (11 loc) · 892 Bytes

SDA Model Performance

Performance results are evaluated on 4-node cluster configured with Intel(R) Xeon(R) Platinum 8358 Scalable processor. For MiniGO, BERT, ResNet, RNN-T, Intel® End-to-End AI Optimization Kit delivered 13.06x, 10.10x, 10.12x and 11.61x training time speedup respecitvely through E2E optimizations. Please refer to corresponding model link for detailed test dataset and test method.

Noted: Optimized lighter models' accuracy are slightly lower: ResNet -5% accuracy, BERT -1% F1 score, RNN-T -1% WER.

Model Training Accuracy Ratio
ResNet 10.12 -5%
BERT 10.10 -1%
RNN-T 11.61 -1%
MiniGo 13.06 0%