Fuck it, I use local LLMs enough, will give this a crack.
Edit: it’s doing 6 paragraphs in 8.2 seconds, the last model I used was doing like 1 paragraph in 12 seconds. Crazy fast in my experience.
How are they to run, how useful are they, and any you can recommend?
Dead simple to run, I use Ollama to run local models and it’s like 3 words to setup from the command line.
Useful is entirely relative. I use mine personally and somewhat professionally, but I only use it to draft text and manually alter it. AI is amazing, but it’s also crap. You gotta work it a bit.
Umm this model from what I can see, I’m using the 8b model and it’s fast to generate, time will tell how good the quality is but I’m impressed after a few minutes play.
8B parameter tag is the distilled llama 3.1 model, which should be great for general writing. 7B is distilled qwen 2.5 math, and 14B is distilled qwen 2.5 (general purpose but good at coding). They have the entire table called out on their huggingface page, which is handy to know which one to use for specific purposes.
The full model is 671B and unfortunately not going to work on most consumer hardwares, so it is still tethered to the cloud for most people.
Also, it being a made in China model, there are some degree of censorship mandated. So depending on use case, this may be a point of consideration, too.
Overall, it’s super cool to see something at this level to be generally available, especially with all the technical details out in the open. Hopefully we’ll see more models with this level of capability become available so there are even more choices and competition.
Also, the release of R1 under the MIT license means that in principle anyone can use R1 to generate synthetic training sets for improving other (non-reasoning) models. This may be a real game changer.
The one fly in the ointment is that Deepseek didn’t deign to share details of their synthetic data generation procedure. But they are already way more transparent than any other non-academic AI lab, so it’s hard to get mad at them over this.
If you want a really simple way to run a variety of local models with a nice UI take a look at https://jan.ai/
so what of its reasoning? can it deduce? can it follow specific logic/equations in mathematical notation or in plain language?
Try it out for yourself: https://chat.deepseek.com/
It can understand LaTeX as well as outputting it. In my limited testing on sample physics problems, it performs pretty well. It also scored 100% on the 2023 A Level maths exam.
interesting, so i guess it can answer questions from any exam
Does it deny Tiananmen square?
Using the 7bn parameter variant:
I’m not sure if this is funny or just sad.
Both
Hahah fuck that’s the funniest most depressing thing ever. Please repost this image I recon in would be a good post.
It’s MIT licensed, so anyone is free to go about decensoring it. There are already “abliterated” (decensored) variants uploaded to huggingface, at least for the distilled models.
This procedure also decensors stuff that western models routinely censor. So ironically these Chinese open source models are giving us the most free speech friendly LLMs around.
I use a dolphin fine tuned meta llama model myself but I will have to compare it to this one.
Have you tried a Tuna Tuned Obama Llama instead?
Asked very plainly, it refuses to answer questions related to it, but it requires very little convincing to talk about it. Much softer censorship than most of the other available models.
How did you convince it? Just curious
