Install locally on MacOS

Running The Web app in your own MacBook

To run the web app in macOS, the instructions are mostly the same as in README.md

In order to be able to run locally, you need a copy of an odinson index and a checkpoint of the scorer.

Necessary files

Small index (100). Download and uncompress: https://drive.google.com/drive/folders/1Wu02GAUNUTQs77uW3Eejk8APKA6TrcZ8?usp=sharing
From clulab's servers, download the file using scp/sftp/rsync /home/rvacareanu/projects/odinsynth/python/logs/odinsynth/version_933/checkpoints/best.ckpt (BERT 8_512): scp river:/home/rvacareanu/projects/odinsynth/python/logs/odinsynth/version_933/checkpoints/best.ckpt . Store this file locally and replace {CKPT_PATH} with the local path below.

Environment

If you're running this on macOS, create a conda environment with the following command:

1conda env create -n odinsynth -f odinsynth_mac.yml

Once installed, activate it with:

1conda activate odinsynth

Now install the node environment to run the front end:

1cd frontend
2npm install

Run the components

There are three components of the demo: Pytorch API, Back end, Front end

Pytorch API

Run the following command replacing {CKPT_PATH} with your local path of the scorer checkpoint refered to in the previous section:

1CHECKPOINT_PATH={CKPT_PATH} ./bash_scripts/start_servers.sh 8000

Back end

Replace {INDEX_PATH} with the location of your local copy:

1sbt -Dodinson.dataDir="{INDEX_PATH}" webapp/run

I struggled making Play! work correctly because homebrew's sbt downloads and uses java version 17. If you have Oracle's JVM version 11 or 8 installed, you can enforce using it with the following variation:

1sbt -java-home "$(/usr/libexec/java_home)" -Dodinson.dataDir="{INDEX_PATH}" webapp/run

Front end

Run the following commands and your browser will open the web interface of the demo:

1cd frontend
2npm start