

- journalctl shows the oom event
- Container limits you can limit container resource usage
- also you can look at container stats to maybe identify the culprit. Or cyclically write the stdout of ps to a file to identify the memory hungry process.
Debian + LVM + Incus :)
*******
Did you type a password?
We need to talk about the horizontal distance between those Ts. (not really :)
Good times.:)
" Can I come with you?"
– little boy in Screamers (1995)
Interesting. What a strange companionship. :)
Hey. I wanted to write that! ;)
What? Ah here it is:
Nvidia CEO Jensen Huang confirmed today that Nvidia will contribute “GPU chiplets” that Intel can place alongside its x86 CPU cores instead of the Arc integrated graphics it develops in-house today.
They could buy AMD Radeon chiplets too, that would be something.:)
data in transit
Yes, but then it is stored unencrypted on github. Ready to be used as training data by github.com/ Microsoft or whoever. Which isn’t bad per se for i. e. public repositories. Just pointing out, that the weakest link in your chain of security measures is the… weakest link.
If you wanted to secure your code, you could store it on-site, behind a firewall, in its own network segment, with encrypted offsite backups. Elliptic curve cypto would help too in this scenario. And MFA. And access restrictions. Many possible measures.
That’s one option but the possibilities are endless.:)
And burn down. And then explode again.
The 🔑 belongs into the locked door, so I can’t leave without it. Oh the horrors of being locked out.
PS: Never leave the house without a pack of paper tissues.
That’s what LLM are made for;
Hence the Name? :)
I mean the automatic speech recognition and transcription capabilities are quite useful. But that’s about it, for me for now.
It could be interesting for frame interpolation in movies at some point maybe, I guess.
I dream of using it for the reliable classification of things. But I haven’t seen it working reliably, yet.
For the creation of abstracts and as a dialog system for information retrieval it doesn’t feel exact/correct / congruent enough to me.
Also: A working business plan to make money with actual AI services has yet to be found. Right now it is playing with a shiny new toy and the expectations and money of their investors. Right now they fail to deliver and the investors might get restless. Selling the business while it is still massively overrated, seems like the only way forward. But that’s just my opinion.
Do it like the mosquitoes do. 🤷♀️
Going to try to make lunch for the week. We’ll see how it goes.
For the entire week? Different dishes? Does it work for you?
The h4 already can be a managed switch itself (2" 2,5gbit + 4*1gbit with the nic addon.) if you want it to be one. Linux as the host OS (VLANs, bridges) - netplan works well for me. Some VMs and containers on top (lxd, incus, some use proxmox) for router/ firewall/ vpn-gateway (opnsense, ipfire,…) and other functionality which you don’t want to run on the host OS directly. The cpu is fast enough to run all your services at once. It all comes down to RAM.
IMO there is not one right way. It all depends on what you want to achieve. Also a lot depends on, whether you want results fast or if you enjoy the tinkering while learning.
rsnapshot is a script for the purpose of repeatedly creating deduplicated copies (hardlinks) for one or more directories. You can chose how many hourly, daily, weekly,… copies you’d like to keep and it removes outdated copies automatically. It wraps rsync and ssh (public key auth) which need to be configured before.