Acronymy Assistant is an interactive tool for composing backronyms.
It was created as an attempt to speed up progress on https://acronymy.net/.
Here is what it looks like:
And here is a quick video explanation of how it works:
Choose a prompt template and edit it to your liking. $WORD will get replaced
by the current word, and $DEF will get replaced by the definition.
If there is no $DEF, then the definition will be added at the end of the prompt.
Higher temperature flattens the probability distributions, i.e. makes Acronymy Assistant more likely to choose weirder tokens.
At each iteration, Acronymy Assistant resamples a random subset of the current definition's words. For each word, replacement_prob is the probability that it will be replace on a given iteration.
If the "sample from model" checkbox is checked for a word, then tokens for that word are chosen from the neural network model. Otherwise, the word is resample uniformly from the word list (by default wordlist.asc). Uniform sampling is kind of like setting the temerature infinitely high, but is additionally constrained to only choose words in the word list.
Set up a virtual environment:
virtualenv venv
source venv/bin/activate
Download the dependecies:
pip install -r requirements.txt
Start the server:
python server.py
Then point a web browser to http://localhost:8080/index.html .
To see commandline options (e.g. specifying which model to use), do this:
python server.py --help