Installation updates #360
Installation updates #360
Conversation
jnwei
commented
Oct 20, 2023
jnwei
commented
- Installation scripts now assumes user has their own version of Anaconda, and provides instructions on how to download anaconda separately.
- Upgrades OpenMM package to use OpenmM 7.7.0 and removes patches
- Updates README instructions
…y/openfold into installation-updates
instead of pip - Adds helper line to automatically prepend conda library to $LD_LIBRARY_PATH
…y/openfold into installation-updates Merging changes from different systems
README.md
Outdated
| To install: | ||
| 1. Clone the repository, e.g. `git clone https://github.com/aqlaboratory/openfold.git` | ||
| 1. From the `openfold` repo: | ||
| - Create an [Anaconda/Mamba](https://docs.anaconda.com/free/anaconda/install/index.html) environment, e.g. `conda env create -n openfold_env` |
There was a problem hiding this comment.
Would say "conda env create -f environment.yml" here
| To activate the environment, run: | ||
|
|
||
| ```bash | ||
| source scripts/activate_conda_env.sh |
There was a problem hiding this comment.
I guess we shouldn't remove these since people will have the old env, but it's kind of confusing that they refer to the old setup
There was a problem hiding this comment.
If we want to keep information/scripts about the old set up, my inclination would be to have a separate section / README for them. These scripts are also more useful if someone is running from a slurm job, so maybe I can move these scripts there and provide some instructions?
|
|
||
| # Grab conda-only packages | ||
| export PATH=lib/conda/bin:$PATH | ||
| lib/conda/bin/python3 -m pip install nvidia-pyindex |
There was a problem hiding this comment.
What was this nvidia pyindex for actually?
There was a problem hiding this comment.
I'm not exactly sure. According to https://github.com/nvidia/tensorflow/#install it's supposed to host NVIDIA wheels that are not part of the regular pypi package, but I didn't find anything that requires it? Especially now that we install flash attention using pip rather than building manually from source?
Maybe @gahdritz would know more.
There was a problem hiding this comment.
With the current environment.yml, I don't this is used anymore, it seems like it was a specific channel for pulling pip packages.
|
|
||
| python setup.py install | ||
|
|
||
| export LD_LIBRARY_PATH=$CONDA_PREFIX/lib:$LD_LIBRARY_PATH |
There was a problem hiding this comment.
Not related to this PR, but could you also add this here to fix that worker issue:
conda env config vars set KMP_AFFINITY=none
There was a problem hiding this comment.
noted. Do we know if the KMP_AFFINITY fix was required for all the systems we tested on, or just a few systems?
There was a problem hiding this comment.
Multiple systems, three different clusters that I know of.
|
|
||
| # The following will prepend conda library environment to $LD_LIBRARY_PATH | ||
| # upon conda library activation | ||
| # conda env config vars set LD_LIBRARY_PATH=$CONDA_PREFIX/lib:$LD_LIBRARY_PATH |
There was a problem hiding this comment.
Not sure about this, I guess $LD_LIBRARY_PATH could be different on diff nodes, this will override the one set on that node with whatever value it had when you first save the env var.
There was a problem hiding this comment.
Yeah, I think you're right, and this is more trouble than it's worth. The first time you deactivate from the environment, it seems to unset the LD_LIBRARY_PATH entirely which can be a bit concerning if you're not expecting that.
…y/openfold into installation-updates
