Skip to content

Latest commit

 

History

History
49 lines (35 loc) · 2.06 KB

internal-instructions.md

File metadata and controls

49 lines (35 loc) · 2.06 KB

Instructions for DSG Hedwig Contributors

Please follow these instructions if you are a graduate student or undergrad research assistant working with the group in the Data Systems Lab and want to run Hedwig on the lab desktop GPU machine (dragon).

If you have trouble / questions with instructions on this page, ping @tuzhucheng on Slack.

PyTorch Environment

We already have a multi-user Conda environment with PyTorch and all other dependencies installed, so you do not need to install anything yourself. However, you can create Conda environments if you need to experiment with different library versions etc.

The multi-user Conda environment is located at /anaconda3/. To use this multi-user environment, just add the following to your .bashrc or configuration file for your favourite shell.

export PATH="/anaconda3/bin:$PATH"
export LIBRARY_PATH="/usr/lib/nvidia-375"

Please also ensure /usr/local/cuda-8.0/lib64 is in the LD_LIBRARY_PATH environment variable if it is not already. If not, you should add it in the .bashrc similar to above.

Please re-login or re-source your shell configuration after .bashrc is updated for the updated environment variables to take effect.

Data and Pre-Trained Models

We use shared cloned versions of the Hedwig-data and Hedwig-models repositories. Instead of making your own cloned copies, you can just create symbolic links to the shared version instead in your own working directory to save disk space. Assuming you want to put Hedwig, Hedwig-data, and Hedwig-models under a directory called Castorini and you are currently in the Castorini directory, you can enter these commands:

ln -s /Hedwig-data Hedwig-data
ln -s /Hedwig-models Hedwig-models

So after you clone Hedwig, you have a directory structure under Castorini that looks like this:

.
├── Hedwig
├── Hedwig-data
└── Hedwig-models

where Hedwig-data and Hedwig-models are actually symbolic links to /Hedwig-data and /Hedwig-models.