Repository to benchmark GPU providing platforms. The code/data in this repository achieves the following:
- Trains a Bi-directional LSTM on sentiment analysis task using twitter data (more than 1.5 million tweets -- processed and contained in the
- Collects various metrics during/after training to help compare various GPU cloud platforms.
- Pins all required library/packages versions, fixes seeds and contains a
Dockerfilewhich can be used to repeat the experiments in a reproducible fashion.
A bidirectional LSTM is trained (using Keras) to perform binary categorization of tweets.
Twitter Sentiment Analysis Dataset containing 1,578,627 classified tweets. This data is split into two files (negative and positive sentiment tweets) which can be found in the
Pull Docker Image
docker pull manneshiva/playground:benchmark-gpu-tfsource-keras
Make sure you create a
results/ folder before running the benchmark.
python benchmark.py --platform aws --epochs 4 --interval 30 --data_ratio 1.0
python benchmark.py -h
Once finished, you can find the results of the benchmark run in