Installation Instructions

This is a python project, but it does have a few non-python dependencies. I have created an environment.yaml file that would make the installation process a little bit easier I hope.

Installing Conda

This one is fairly straightforward, you can follow the instructions here after running the installation script you should be able to activate/deactivate conda environments. If you are installing this under HPC, I suggest that you move the location of your conda cache to a different location. You can do that by following the instructions here, the other option is that you can create a symbolic link to your .cache, .singularity and .conda folders in your ~ where the actual folders are in a partition with more storage.

Installing Benchmate

The project comes with a conda enviroment.yaml file that has all the dependencies listed. Keep in mind that this project is still under heavy development and the dependencies might change and that might result in older installations to break.

First you will need to clone the repository using git.

# clone the repository
git clone https://github.com/ccmbioinfo/ccm_benchmate

# enter benchmate director
cd ccm_benchmate

Then you can create the conda environment and install all the dependencies with a single command.

conda create env -f environmemt.yaml

conda activate benchmate

This will install all the dependencies including a postgres server. This is needed for using the project instance an all the related modules, if you are not interested in that you can comment out that section in the yaml file.

After installing dependencies you will need to install benchmate itself.

pip install . 

If you are using postgres

Benchmate will not create a database for you. You will need to do that as the pg admin.

Start the database server:

You will need to start the database server, postgres comes with many reasonable defaults but if you need to change them please check their documentation. The default port for postgres is 5432.

To start the database server you will need to specify a folder and a logfile. The directory needs to be empty.

# intialize the process, if <database_dir> does not exist you need to create it
initdb -D <database_dir>

#start the server
pg_ctl -D <database_dir> -l <database_dir>/logfile start

With this you will have posgres server running, then you will need to log into the databse and create a new one.

psql -d template1 #this one is a small template that comes with db installation

Create the database and user

CREATE ROLE <username> WITH LOGIN SUPERUSER PASSWORD `password`; 

CREATE DATABASE <your_db_name> OWNER <username>

A few of the modules (literature and molecule) depend on postgres extensions we need to activate them.

create extension if not exists rdkit; 

create extension if not exists vector;

Then we should be good to go.

A Quick note about model selection

You can see in the config.yaml file that we have made some decisions about which models to use as a default in the package. These decisions were made with a couple of important considerations in ming.

1. The models need to be robust and actually follow instructions (not every instruction tuned model does that or does it reliably)
2. They need to be small enough to be run in <40GB of VRAM (you can probably run benchmate no problem with a GTX5090 that you can buy at a computer hardware store)
3. They are fast(-ish), this was the main consideration for choosing a static model for semantic chunking.

Containers

We are working on creating docker containers and a docker_compose.yaml file to make this a less painful process. Once those developments are done this documentation will be updated.

Please create an issue with all the error messages if you run into problems.