sntx

It's affected by the write-hole phenomenon. In BTRFS case that can mean that perfectly good old data might corrupt without any notice.

~/Documents/projects/<YYYY>-<MM>-<DD>_<name>

Fascinating idea!

I'm using rustic, a lock-free rust-written drop-in-replacement of restic, which (I'm referring to restic and therefore in extension to rustic) supports always-encrypted, deduplicating, compressed and easy backups without you needing to worry about whether to do a full- or incremental-backup.

All my machines run hourly backups of all mounted partitions to an append-only repo at borgbase. I have a file with ignore pattern globs to skip unwanted files and dirs (i.e.: **/.cache).

While I think borgbase is ok, ther're just using hetzner storage boxes in the background, which are cheaper if you use them directly. I'm thinking of migrating my backups to a handfull of homelabs from trusted friends and family instead.

The backups have a randomized delay of 5m and typically take about 8-9s each (unless big new files need to be uploaded). They are triggered by persistent systemd-timers.

The backups have been running across my laptop, pc and server for about 6 months now and I'm at ~380 GiB storage usage total.

I've mounted backup snapshots on multiple occasions already to either get an old version of a file, or restore it entirely.

There is a tool called redu which is like ncdu but works on restic/rustic repos. This makes it easy to identify which files blow up your backup size.

That woul've been: Minetest Immortal

Thanks for the writeup! So far I've been using ollama, but I'm always open for trying out alternatives. To be honest, it seems I was oblivious to the existence of alternatives.

Your post is suggesting that the same models with the same parameters generate different result when run on different backends?

I can see how the backend would have an influence hanfling concurrent api calls, ram/vram efficiency, supported hardware/drivers and general speed.

But going as far as having different context windows and quality degrading issues is news to me.

Is there an inherent benefit for using NVLINK? Should I specifically try out Aprodite over the other recommendations when having 2x 3090 with NVLINK available?

If you connect to the network and open firefox, it will display a toast to open the corresponding captive portals page. You can then login through that. Given that your VPN isn't blocking unencrypted connections etc.

Extrapolation of partial knowledge warning

I assume the network advertises a captive portals url and identifies you based on your MAC address.

The config is server-side (router).

yes: sntx.space, check out the spurce button in the bottom right corner.

I'm building/running it the homebrewed-unconventional route. That is I have just a bit of html/css and other files I want to serve, then I use nix to build that into a usable website and serve it on one of my homelab machines via nginx. That is made available through a VPS running HA-Proxy and its public IP. The Nebula overlay network (VPN) connects the two machines.

What do you use it for? How's the daily-driver experience?

Is something like this defined in a standard somewhere?

RFC3389 ?