DLProf Examples on Summit

This repo consists of usage examples of Nvidia DLProf on the Summit supercomputer.

DLProf is a tool for profiling deep learning models to help data scientists understand and improve performance of their models visually via Tensorboard or by analyzing text reports.

Quickstart

• use DLProf on Summit

module use /sw/aaims/summit/modulefiles
module load dlprof

• refer this blog for more details.

PyTorch Example

git clone --recursive https://github.com/at-aaims/dlprof-examples
cd dlprof-examples/DeepLearningExamples
git apply ../pytorch/ConvNets.patch
cd ../pytorch 
bsub prof.lsf

Visualize Output

• install tensorboard plugin (for x86 only)

pip install nvidia-pyindex
pip install nvidia-tensorboard
pip install nvidia-tensorboard-plugin-dlprof

• use pre-installed env on Andes

module load python
source activate /gpfs/alpine/world-shared/stf011/junqi/dlprof-env
tensorboard --logdir /gpfs/alpine/world-shared/stf011/junqi/dlprof-env/event_files --host localhost

• port forward to local machine

http://localhost:6006