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
To cite 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 %}