Open trip54654 opened 8 months ago
this tbh
I made a quick & dirty port to node.js. The userscript is mostly the same, except that it doesn't load tensorflow and doesn't inline the weights. It talks to the nodejs server with HTTP requests. Hoping this will fix my firefox memory leak problems.
Does anyone know answers to these questions:
It's pretty clear that this solves my memory leak problems.
This is why the extension version was popular. It was much lighter while also quicker. However it seems it was abandoned.
Problem: the model is huge and uses a lot of memory. The model is probably duplicated for each tab.
Solution: create the option to use an external captcha solver server on localhost. The server could be a simple nodejs or python script that serves HTTP on localhost and gets accessed by the script.
Maybe this script is even responsible for the Firefox higher memory usage I'm observing:
But I don't know how to debug this.