I’ve been playing with the largest models I can get running and have been using Librewolf or Firefox, but these use several gigabytes of system memory. What options exist that have less overhead? I’m mostly looking at maximizing the model training potential as I’m learning. The obvious solution is python in a terminal, but I need a hiking trail not free solo rock climbing.
Elinks or lynx
Absolutely right. I just tried it on the browsers installed on my system, loading this page:
Firefox: 560MiB
Epiphany (GNOME Web): 226MiB
elinks: 16MiB
lynx: 14MiBLooks like lynx is the winner
(Sidenote: This isn’t really a fair fight for Firefox since it’s my daily driver, with extensions installed and a bunch of stuff cached. I’m guessing even a fresh install wouldn’t get below 300MiB, though)
deleted by creator
Something webkit based probably. Gnome web is probably the most accessible of these.
Not what you’re asking for, but how about putting the web browser and the page rendering on a different machine? This way your main machine can focus on calculating.
Edit: If the pages are super simple, there’s “web browsers” which do work on the command line which can render simple pages in a very crude way.
This is kind of what I was thinking. I have a $5/mo VPS running Selenoid. All it does is take incoming requests to simulate a real user clicking on some stuff.
Basically I run a website in Ruby on Rails that has to talk to some API’s. Unfortunately the industry that app works in is very behind with tech, so I make do with simulating a user visiting some portals en lieu of actual API calls. It’s great because the resource-constrained containers don’t have to power up an entire web browser in background jobs, though running these tasks as long-running background jobs presents other issues.
There’s a reason these browsers use that much memory. Something in living there and that’s not just overhead. You can’t realistically reduce that by a reasonable amount by just using another browser while retaining functionality.
