My intuition:

• There’re “genuine” instances of hapax legomena which probably have some semantic sense, e.g. a rare concept, a wordplay, an artistic invention, an ancient inside joke.
• There’s various noise because somebody let their cat on the keyboard, because OCR software failed in one small spot, because somebody was copying data using a noisy channel without error correction, because somebody had a headache and couldn’t be bothered, because whatever.
• Once a dataset is too big to be manually reviewed by experts, the amount of general noise is far far far larger than what you’re looking for. At the same time you can’t differentiate between the two using statistics alone. And if it was manually reviewed, the experts have probably published their findings, or at least told a few colleagues.
• Transformers are VERY data-hungry. They need enormous datasets.

So I don’t think this approach will help you a lot even for finding words and phrases. And everything I’ve said can be extended to semantic noise too, so your extended question also seems a hopeless endeavour when approached specifically with LLMs or big data analysis of text.

No IPA notation? ⸨I’m somewhat disappointed⸩

It would. But it’s a good option when you have computationally heavy tasks and communication is relatively light.

Once configured, Tor Hidden Services also just work (you may need to use some fresh bridges in certain countries if ISPs block Tor there though). You don’t have to trust any specific third party in this case.

If your CPU isn’t ancient, it’s mostly about memory speed. VRAM is very fast, DDR5 RAM is reasonably fast, swap is slow even on a modern SSD.

8x7B is mixtral, yeah.

Mostly via terminal, yeah. It’s convenient when you’re used to it - I am.

Let’s see, my inference speed now is:

• ~60-65 tok/s for a 8B model in Q_5_K/Q6_K (entirely in VRAM);
• ~36 tok/s for a 14B model in Q6_K (entirely in VRAM);
• ~4.5 tok/s for a 35B model in Q5_K_M (16/41 layers in VRAM);
• ~12.5 tok/s for a 8x7B model in Q4_K_M (18/33 layers in VRAM);
• ~4.5 tok/s for a 70B model in Q2_K (44/81 layers in VRAM);
• ~2.5 tok/s for a 70B model in Q3_K_L (28/81 layers in VRAM).

As of quality, I try to avoid quantisation below Q5 or at least Q4. I also don’t see any point in using Q8/f16/f32 - the difference with Q6 is minimal. Other than that, it really depends on the model - for instance, llama-3 8B is smarter than many older 30B+ models.

Have been using llama.cpp, whisper.cpp, Stable Diffusion for a long while (most often the first one). My “hub” is a collection of bash scripts and a ssh server running.

I typically use LLMs for translation, interactive technical troubleshooting, advice on obscure topics, sometimes coding, sometimes mathematics (though local models are mostly terrible for this), sometimes just talking. Also music generation with ChatMusician.

I use the hardware I already have - a 16GB AMD card (using ROCm) and some DDR5 RAM. ROCm might be tricky to set up for various libraries and inference engines, but then it just works. I don’t rent hardware - don’t want any data to leave my machine.

My use isn’t intensive enough to warrant measuring energy costs.

There’s a recent algorithm using somewhat similar ideas for approximate counting of unique objects in a stream with constant memory:

https://www.quantamagazine.org/computer-scientists-invent-an-efficient-new-way-to-count-20240516/

Disabling root login and password auth, using a non-standard port and updating regularly works for me for this exact use case.

Alas, their IP requirements are too much for me.

Anyway, is there much of xenharmonic sheet music there? Take, for example, Easley Blackwood (but not his books).

Downloading there is straightforward: look at network requests, redownload svg’s of individual pages with wget and reassemble those into a pdf. I did that today and the resulting quality wasn’t exactly low - though I didn’t examine it too closely. Readability was perfect.

Probably could be automated, but I’m not bothered enough to do so yet.

I have a MediaWiki instance on my laptop (I’ve found the features of all other wikis/mindmaps/knowledge databases decisively insufficient after having a taste of MW templates, Semantic MediaWiki and Scribunto).

Also some smaller things like pihole-standalone, Jellyfin and dictd.