Technically this isn’t the case. These are Deepseek distilled with Qwen and Llama depending on the weight, so they are no longer Deepseek.
@YarHarSuperstar@lemmy.world this is the official, extremely large, one. https://huggingface.co/deepseek-ai/DeepSeek-R1
Let’s give credit where it’s due: https://github.com/maybenot-io/maybenot
The Maybenot Framework (FOSS) is how Mulvad pulls this off, and if you run your own VPN you can use this too! Mulvad is a contributor (and funder), so good on them.
Edit: for those interested, Mulvad’s client is a fork of Wireguard with Maybenot incorporated as a submodule. Cool stuff: https://github.com/mullvad/wireguard-go
Since the other answer is desktop use, if in Linux your best bet is smartctl.
Every time they post on Mastodon they should make a post on Twitter that says they posted there until they get banned.
Credits aside, these dont seem like bad numbers when compared to spend/investment. Tesla management got what they wanted.
There are a bunch of things that I think are value-destroying for me to talk about, so I’m not going to talk about those.
Mmhmm.
Fast for the benchmarks. “We’ll make it slower and safer later.”
My instance of one now federates with your instance of one.
For the record, not everyone thinks ethics in licensing is a bad idea.
And for those using Matrix, it stops you from having to use Google notifications.
Works great for me, too.
Hmm, I always adjust volume with the keyboard, so have never run into this.
The scenario you describe is actually represented in a great free book from Cory Doctorow: https://boingboing.net/2009/10/28/makers-my-new-novel.html
Today is the launch of my new novel, Makers, a book about people who hack hardware, business-models, and living arrangements to discover ways of staying alive and happy even when the economy is falling down the toilet.
I’ve had pretty good luck with everything except Google Voice, since it requires Electron and the Docker containers don’t use it. But I’ve had good results withbridging Signal, Discord, and WhatsApp in my local install.
So use a Matrix bridge? Or Beeper (which is a Matrix bridge)?
Personally the part I like is that it’s not meta. Unfortunately if 8b is based on llama, there could be meta censorship baked in that we simply don’t know about.
Just remember, Ollama’s version of 8b models is not the same as the original on Huggingface. There’s a reason it’s a much smaller file size. That being said my understanding is the quant is good.
I guess don’t visit the new https://www.whitehouse.gov/ then.
Oh nice, when Ollama first had this up I don’t think the 671b model was there. I stand corrected on that!