jash271 / detoxify Goto Github PK

Current ML model is saved in drive.

While running the application on cloud major problems which one can faces are:

• downloading the model hosted on cloud
• starting the app with uvicorn. Easy though, but a bash script will be helpful.

Consider this as child issue of #3

Create a Client-Side chrome extension - Much like how Grammarly is, where the extension can be overlaid on any textbox(like Grammarly) and show stats of what has been written in that particular textbox. This extension should work on most of the sites and get overlaid on those text boxes to show the stats for the text that has been written

The stats that you would get from backend would be something like this, how these stats are to be shown is up to your creativity:

{
"result": "(MultiCategory tensor([1., 1., 1., 0., 1., 0.]), tensor([1., 1., 1., 0., 1., 0.]), tensor([0.9999, 0.9452, 1.0000, 0.0019, 0.9661, 0.0018]))",
"list": [
0.9999,
0.9452,
1,
0.0019,
0.9661,
0.0018
],
"dict": {
"toxic": 0.9999,
"severe_toxic": 0.9452,
"obscene": 1,
"insult": 0.9661
}

We are interested in the final dictionary being overlaid on the textbox

Apparently, due to fastai, deployment of the API on Heroku has been a headache, would appreciate if anyone can solve this deployment issue so the frontend team's work becomes easier

The current project doesn't have any healthcheck API endpoint to know if the API is up and running or not.

• This issue proposes a healthcheck endpoint.
• This issue proposes to add few more allowed origin hosts to make it work on cloud instances. [already done in origins]

Issue can be conuted as a child issue of #3.

Issue to propose adding requirements.txt and .gitignore for easy installation.

This particualr line in the code

res = learn.predict(k)
leads to this : NumExpr defaulting to 8 threads

and the result isn't properly displayed,so had to use regex and work around to get the toxicity score

Would appreciate if anyone can fix this or has a better work around