Wait, just the client? I thought the madlad ran the model on their phone!
apparently not. it seems they are refering to the official bs deepseek ui for ur phone. running it on your phone fr is super cool! Imma try that out now - with the smol 1.5B model
Good luck ! You’ll need it to run it without a GPU…
i kno! i'm already running a smol llama model on the phone, and yeaaaa that's a 2 token per second speed and it makes the phone lag like crazy... but it works!
currently i'm doing this with termux and ollama, but if there's some better foss way to run it, i'd be totally happy to use that instead <3
i think termux is probably already the best way to go, it ensures linux-like flexibility, i guess. but yeah, properly wiring it up, with a nice Graphical User Interface, would be nice, i guess.
Edit: now that i think about it, i guess running it on some server that you rent, is maybe better, because then you can access that chat log from everywhere, and also, it doesn't drain your battery so much. But then, you need to rent a server, so, idk.
Edit again: Actually, somebody should hook up the DeepSeek chatbot to Matrix chat, so you can message it directly through your favorite messaging protocol/app.
edit again: (one hour later) i tried setting up deepseek 8b model on my rented server, but it doesn't have enough RAM. i tried adding swap space, but it doesn't let me. I figured out that you can't easily add swap space in a container. somehow, there seems to be a reason to that. too tired to explore further. whatever.
big sad :(
wish it would be nice and easi to do stuff like this - yea hosting it somewhere is probably best for ur moni and phone.
actually i think it kinda is nice and easy to do, i'm just too lazy/cheap to rent a server with 8GB of RAM, even though it would only cost $15/month or sth.
it would also be super slow, u usually want a GPU for LLM inference.. but u already know this, u are Gandald der zwölfte after all <3
wait what's bad about it?
~i~ ~installed~ ~llama~ ~on~ ~my~ ~phone...~
A US senator has introduced a bill trying to criminalize it with a 20 year prison sentence.
Also, the US and El Salvador are doing a prisoner deportation deal.
I'm not making this up.
Capitalism hates competition
If it's running locally nothing.
If you install the actual deepseek app, you might as well post your prompts openly on social media.
oh yea, installing the app is super dum. your point is very good.
It doesn't require any permissions.
Meanwhile Facebook, Google, Snap Chat, fucking TEAMS, require you to give them access to your camera and microphone 24/7
But yeah app bad cuz china
its all about peeps sharing their personal info on these chatbots. Many peeps do, so I guess it kinda makes sense.
I normally use it to help me with my English translations here on lemmy (I can more or less get by but I prefer it to be correct), so I already do it.
You can also use DeepL for that, which I’m willing to bet uses a lot less energy and is a very reliable translation service.
Yes and I use it more often, but sometimes a literal translation is not enough, I know enough English to notice when something is weird and in those cases I go to chatgpt or deepseek who understand what you mean and its context and give you the same thing and with the same tone in another language. I don't use it all the time, but it greatly improves the result.
So...this whole thing is about hiding lewd prompts?
nono, the whole thing is about some people putting personal info into these chatbots.
and even if not, they are guaranteed to train their newer models on the requests and generated responses.
if ur putting personal info, running in locally/privately is kinda a must, if u care about security at all.
i think peeps try lewd prompts once, then find out it doesn't work, and then give up. (they don't know about huggingface)
Why would anyone....you know, nevermind 😂
I would never install Chinese spyware on my Samsung!
/s
Be sure to follow the rule before you head out.
Rule: You must post before you leave.
