Docker configs, miner.py revisions, pytorch-lighting and lighting-hivemind poc

[Vectorrent]

This code implements a basic Docker configuration that will allow peers to run this project inside of Docker, with GPU passthrough.

System requires Docker, Docker Compose, and Nvidia Container Toolkit.

The basic commands:

docker compose build

# push newly-built image to container registry
docker compose push 

docker compose up

# to enter the container environment interactively
docker compose exec -it hivetrain /bin/bash 

Hope it helps. Feel free to let me know if you guys have any questions!

[Vectorrent]

The new neurons/miner-z.py script is functional, though it is very badly-designed.

Easiest way to use it is to configure the Docker stuff (see above), then create a .env file with INITIAL_PEERS set. See the example.env file. My current, training peer ID is:

INITIAL_PEERS="/p2p/12D3KooWCvMCCJDHQ7d9pfqqkxAPD6AZdAbcXPd1d9pWvQWDpqBi"

If you want to run on bare metal (i.e. without Docker), look at Dockerfile as a reference for dependency installs. Then, look at entrypoint.sh for the startup command. You need to pass my peer ID:

python3 miner-z.py --initial_peers "/p2p/12D3KooWCvMCCJDHQ7d9pfqqkxAPD6AZdAbcXPd1d9pWvQWDpqBi"

This is far from optimal. Currently, TQDM is broken and needs work.