RL-Scope: Cross-Stack Profiling for Deep Reinforcement Learning Workloads

RL-Scope collects cross-stack profiling information (CUDA API time, GPU kernel time, ML backend time, etc.), and provides a breakdown of CPU/GPU execution time.

RL-Scope's complete documentation can be found here: https://rl-scope.readthedocs.io/en/latest/index.html

Here are some convenient links to common parts of the documentation:

• Installation
• RL-Scope artifact evaluation
• Interactive "Getting Started" notebook on Google Colab
• Docker development environment
• YouTube: Lightning talk - 5 minutes
• YouTube: Full talk - 17 minutes
• RL-Scope paper

Paper

For convenience, you can find our paper on arxiv: https://arxiv.org/abs/2102.04285

When citing RL-Scope, please cite our MLSys 2021 publication:

{% raw %}

@inproceedings{gleeson2021rlscope,
 author = {Gleeson, James and Krishnan, Srivatsan and Gabel, Moshe and Janapa Reddi, Vijay and de Lara, Eyal and Pekhimenko, Gennady},
 booktitle = {Proceedings of Machine Learning and Systems},
 title = {{RL-Scope:} Cross-Stack Profiling for Deep Reinforcement Learning Workloads},
 year = {2021}
}

{% endraw %}

Videos

YouTube videos recordings describing RL-Scope are available here:

• Lightning talk - 5 minutes
• Full talk - 17 minutes