How’s your stuff doing? Unplanned interruptions or achieving uptime records?
I’m currently sailing rather smooth. Most of my stuff is migrated to Komodo, there will stay some exceptions and I only have to migrate Lemmy itself I think. Of course that’s when I found a potential replacement but I’ll let it sit for a while before touching it again. Enjoying the occasional Merge Request notification from the Renovate Bot and knowing my stuff is mostly up to date.
I’m thinking about setting up some kind of Wiki for my other niche hobby (Netrunner LCG) lore as there’s a fandom one that most people avoid touching and updating but since I likely won’t have time to start writing some articles on my own as a kickoff I’m hesitant. Also not sure which wiki I’d choose as well.
I recently switched my phone from Android to GrapheneOS and now rely even more on my selfhosted services. Immich is such a great project. Still gotta figure out my music collection though, since switching from YT Music to Jellyfin. Most of it is sorted by date of purchase, because that worked best with my DJ workflow. Now I gotta bring it over to a folder structure that works for jellyfin. It seems like the answer is musicbrainz Picard, but I gotta figure out how to configure it.
Also been thinking about some AI ideas I’d like to try, but I have zero intention getting involved with openai, meta, google or whoever the fuck. So self hosting it is. But on what hardware? Option 1 seems to be to get some professional server board, CPU, ram and start with one RTX3090 and go from there with the option to hook up more GPUs. But a setup like that sounds like it would cost some serious money in electricity. Option 2 seems to be a Rzyen AI Max+ 395, configured with a fuckton of ram, available to the whole apu and as suchs usable for memory hungry models. This seems to be much much more power efficient. But its all integrated and I couldn’t swap out components or upgrade in the future. Leaning towara option 2 atm, but maybe I’ll just wait a bit longer and see what else comes up in the coming months.
I ended up using navidrome for my music to take advantage of the subsonic API, which has been phenomenal 🙂
I’ve just finished ripping about 1300 CDs. I used Jellyfin for a bit since I already had it set up for my video library, but I wasn’t happy with the Android options and it was pretty basic.
Navidrome is a fucking TREAT. Paired with Symfonium, I’m finally enjoying my personal music collection regularly again.
As for tagging, OP, while I get why people like Picard, it doesn’t always work with how I like to do things. I put everything into a music folder on my desktop, use Mp3Tag to retrieve metadata, edit what I need, and make sure the artwork is decent and sized where I want it. Then I use the tag > filename to organize and move them to my NAS.
I’ve found Navidrome to be quite capable of handling large music collections. I was worried in the beginning. It sips resources. When I fire it up and listen remotely, I watch the CPU and RAM. It barely moves the needle. Very happy with it.
Nice… I use ytdl-sub for downloading music, highly recommend it. You can write tag metadata but if you want embedded stuff I’d recommend trying beets. Running both as a user whose primary group matches Jellyfin is a must if you want stuff saved next to the video files… The dev is also very active.
I just installed Ollama and use gemma3 for now. I wanted to use dolphin-mixtral but holy crap it wants more RAM than my entire setup
This is basically what I’ve found with self hosted AI. I just don’t have the equipment for it. Would love to be able to host a selfcontained LLM, but alas, as you say, it eats up resources. FEED ME MAURICE!