plex

README

PLEX

PLEX = PLEX Lab EXchange.

Installing the client

First, install the client by running

source <(curl -sSL https://raw.githubusercontent.com/labdao/plex/main/install.sh)

The installer may ask for your password at some point.

When the installer is complete, next update the executable's permissions.

chmod +x ./plex

Running the client

Once the client is installed, you can run the following command in the newly-created plex folder to run equibind.

./plex -app equibind -input-dir ./testdata/binding/abl

Building PLEX from source

git clone https://github.com/labdao/plex
cd plex
go build

Running a node

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 IFPS 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

Notes

To download large bacalhau results the below command may need ran

sudo sysctl -w net.core.rmem_max=2500000