Install using docker

Requirements

Docker
docker-compose

On a Mac, simply use Docker Desktop:

1brew install --cask docker

On Linux, follow the official distro-specific instructions.

0. Copy the data locally

We'll need an index to run things locally. The index should be stored under the ./data from the project root under the odinson directory. Your directory structure for ./data (and directory names!) should mirror this output of tree -L 2:

1data
2└── odinson
3    ├── docs
4    └── index

Options

100K sample of UMBC: /data/nlp/corpora/umbc/umbc_sample_100k.tar.gz

1. Build docker images

Option a: build for ARM64 machines

If you're running things on a Mac M1 or an ARM64-based machine, ...

1./build-images arm64

Option b: build for Intel/AMD machines

If you're running things on an Intel or AMD machine, ...

1./build-images amd64

This will produce the following images:

1odinson/odinsynth-ui:latest
2odinson/odinsynth-backend:latest

You can verify this by running docker images

2. Port forwarding for GPU-accelerated inference

Rather than run the program synthesis service locally on the CPU, we'll use a GPU-accelerated service running on a remote machine (ex. rogue). In another terminal/tmux pane, run the following (replacing <remoteport> with the appropriate port on the remote):

1ssh -L 8000:localhost:<remoteport> rogue

If you've done this correctly, running the following command in another window should return an array with three numbers:

1curl -X POST -H "Content-Type: application/json" -d '{"sentences":[["San","Jose"]],"specs":[{"docId":"test-doc","sentId":0,"start":0,"end":2}],"patterns":["□ □","□?","□*","□+"],"current_pattern":"□"}' 127.0.0.1:8000/score

You should see a result like the following:

1[0.9998440742492676,0.00011766255192924291,0.00011426166020100936,0.00012613831495400518]

3. Launch the docker services

1docker-compose up