star9daisy/hml-env repository overview

⁠HML ENV

⁠Introduction

The hml-env provides a development environment for interdisciplinary research between high-energy physics and artificial intelligence. Based on the included software and libraries, it is divided into three environments:

1. base: Provides the fundamental software for phenomenological research in high-energy physics and can generate simulated collision events.

2. lite: Builds upon base by adding support for TensorFlow 2.16.

3. dev: Builds upon lite by adding support for a complete machine learning framework, enabling comprehensive research.

⁠Configurations

• lite and dev are built based on the image nvidia/cuda:12.2.2-cudnn8-devel-ubuntu22.04. Before starting the container, you need to install the NVIDIA Container Toolkit. Please refer to here⁠ for installation instructions.

• Delphes is installed separately for performance considerations during compilation. The software installed by MG5 has also undergone some minor modifications to utilize all CPU cores during compilation.

• All three machine learning frameworks support GPU and can be used as backends for Keras versions 3 and above.

• The memory growth of TensorFlow and Jax is limited to prevent a single program from occupying the entire GPU memory at once.

⁠Use cases

1. Launching the Container and Generating Events with Madgraph5:

   docker run -it --rm star9daisy/hml-env:3.0.0-base
   

   Copy

   After entering the container, type mg5_aMC to open the Madgraph5 command-line interface.

2. Starting the Container in the Background and Attaching with VSCode:

   docker run -itd --gpus all star9daisy/hml-env:3.0.0-dev
   

   Copy

   After starting, attach VSCode to the running container, allowing you to open the container as if it were a local directory.

3. The image includes an SSH service, and you can add port mappings and passwords when starting the container:

   docker run -itd --gpus all --publish <port>:22 --env PASSWORD=<password> star9daisy/hml-env:3.0.0-dev
   

   Copy

   When this container is running on a remote server, doing so can save a step: changing from local → remote → container to local → container. Use the following command to SSH directly into the container:

   ssh -p <port> root@<remote server ip>
   

   Copy