⚡ Run highly reproducible scientific applications on top of a decentralised compute and storage network. ⚡
PLEX is a simple client for distributed computation.
- 🌎 Build once, run anywhere: PLEX is using distributed compute and storage to run containers on a public network. Need GPUs? We got you covered.
- 🔍 Content-addressed by default: Every file processed by PLEX has a deterministic address based on its content. Keep track of your files and always share the right results with other scientists.
- 🪙 Ownernship tracking built-in Every compute event on PLEX is mintable as an on-chain token that grants the holder rights over the newly generated data.
- 🔗 Strictly composable: Every tool in PLEX has declared inputs and outputs. Plugging together tools by other authors should be easy.
PLEX is based on Bacalhau, IPFS, and inspired by the Common Workflow Language.
- Install Plex with pip
pip install PlexLabExchange
- Run Plex example in a Python file, notebook or REPL
from plex import plex_run
io_json_cid, io_json_local_filepath = plex_run('QmWdKXmSz1p3zGfHmwBb5FHCS7skc4ryEA97pPVxJCT5Wx')
1 . Install the client
Mac/Linux users open terminal and run
source <(curl -sSL https://raw.githubusercontent.com/labdao/plex/main/install.sh)
Windows users open terminal as an adminstrator and run
Invoke-Expression (Invoke-WebRequest -Uri "https://raw.githubusercontent.com/labdao/plex/main/install.ps1" -UseBasicParsing).Content
- Submit an example PLEX job
./plex create -t tools/equibind.json -i testdata/binding/abl --autoRun=True
-
Read the docs to learn how to use PLEX with your own data and tools
-
Request Access to our VIP Jupyter Hub Enviroment and NFT Testnet Minting. VIP Beta Access Form
- 🧬 run PLEX to design proteins with colabfold and RFDiffusion
- 💊 run PLEX to run small molecule docking with equibind and diffdock
- 🐋 configure your containerised tool to run on PLEX
git clone https://github.com/labdao/plex
cd plex
go build
This is a script for setting up a compute instance to run LabDAO jobs. Requires linux OS with Nvidia GPU.
Tested on Ubuntu 20.04 LTS with Nvidia T4, V100, and A10 GPUs (AWS G4, P3, and G5 instance types)
The install script sets up Docker, Nvidia Drivers, Nvidia Container Toolkit, and IPFS
curl -sL https://raw.githubusercontent.com/labdao/plex/main/scripts/provide-compute.sh | bash && newgrp docker
After the script run the following command in a separate terminal to start a Bacalhau server to accept jobs.
ipfs daemon
Once the daemon is running, configure the Bacalhau node based on the addresses used by the IPFS node.
ipfs id
# copy the ip4 tcp output and change port 4001 to 5001 then export
export IPFS_CONNECT=/ip4/127.0.0.1/tcp/5001/p2p/<your id goes here>
# example: export IPFS_CONNECT=/ip4/127.0.0.1/tcp/5001/p2p/12D3KooWPH1BpPfNXwkf778GMP2H5z7pwjKVQFnA5NS3DngU7pxG
LOG_LEVEL=debug bacalhau serve --job-selection-accept-networked --limit-total-gpu 1 --limit-total-memory 12gb --ipfs-connect $IPFS_CONNECT
To download large bacalhau results the below command may need ran
sudo sysctl -w net.core.rmem_max=2500000
PRs are welcome! Please consider our Contribute Guidelines when joining.
From time to time, we also post help-wanted
bounty issues - please consider our Bounty Policy when engaging with LabDAO.