READ ME File For 'Dataset for "Dynamic Transformer for Efficient Machine Translation on Embedded Devices"'

Dataset DOI:  https://doi.org/10.5258/SOTON/D1908

ReadMe Author: Hishan Parry, University of Southampton

This dataset supports the publication:
AUTHORS: Hishan Parry, Lei Xun, Amin Sabet , Jia Bi , Jonathon Hare , Geoff V. Merrett
TITLE: Dynamic Transformer for Efficient Machine Translation on Embedded Devices
CONFERENCE: 3rd ACM/IEEE Workshop on Machine Learning for CAD (MLCAD)
PAPER DOI IF KNOWN:


This dataset contains:
Data for Figure.1,Figure.3,Figure.4,Figure.5

The figures are as follows:

Figure.1
Experimental results to showing how latency constraints set during design time optimization are not met at run-time for a Transformer model on the Jetson Nano, due to a varying availability of computing resources or power.

Figure.3 
BLEU Score vs experimentally measured Latency on the Jetson Nano GPU, showing how the number of decoder layers affect translation performance of the sampled SubTransformer model.

Figure.4  
Validation Loss vs experimentally measured Latency on the Jetson Nano GPU, showing how number of decoder layers affect inherited validation loss of the sampled SubTransformer model.

Figure.5 
Comparing BLEU scores of operating points for the reduced SuperTransformer design space compared to the original design space.

Date of data collection: October 2020 - May 2021

Information about geographic location of data collection: Southampton, UK

Licence: CC BY

Related projects: International Centre for Spatial Computation

Date that the file was created: July 2021