opendilab / DI-hpc

OpenDILab RL HPC OP Lib, including CUDA and Triton kernel

DI-HPC: Decision Intelligence - High Performance Computation

DI-HPC is an acceleration operator component for general algorithm modules in reinforcement learning algorithms, such as GAE, n-step TD and LSTM, etc. The operators support forward and backward propagation, and can be used in training, data collection, and test modules.

Requirements

Setting 1

• CUDA 9.2
• PyTorch 1.5 (recommend)
• python 3.6 or python 3.7 or python3.8
• Linux Platform

Setting 2

• CUDA 9.0
• gcc 5.4.0
• PyTorch 1.1.0
• python 3.6 or python 3.7
• Linux Platform

Note: We recommend that DI-HPC and DI-Engine share the same environment, and it should be fine with PyTorch from 1.1.0 to 1.10.0.

Quick Start

Install from whl

The easiest way to get DI-HPC is to use pip, and you can get .whl from

• di_hpc_rll-0.0.2-cp36-cp36m-linux_x86_64.whl
• di_hpc_rll-0.0.2-cp37-cp37m-linux_x86_64.whl
• di_hpc_rll-0.0.2-cp38-cp38-linux_x86_64.whl

and then call

$ pip install <YOUR_WHL>

Install from source code

Alternatively you can install latest DI-HPC from git master branch:

$ python3 setup.py install

Run on Linux

You will get benchmark result by following commands:

$ python3 tests/test_gae.py

TODO

• [] Trition Kernel for Reinfocement Learning

Feedback and Contribution

• File an issue on Github
• Discuss on DI-engine's (also for DI-hpc) discord server
• Contact our email ([e-mail])

We appreciate all the feedbacks and contributions to improve DI-engine, both algorithms and system designs. And CONTRIBUTING.md offers some necessary information.

License

DI-hpc released under the Apache 2.0 license.