AtHeartEngineer @ AtHeartEngineer @lemmy.world

Aspiring polymath. Applied R&D @ Privacy and Scaling Explorations #make

---

Read more

Posts

9

Comments

432

Joined

2 yr. ago

Moderating

---

Try endeavoros, it's an opinionated arch with a simple installer

OPM also houses all the clearance applications and associated investigation paperwork. Great..

Prohibition of technology import is insane, how fucking moronic.

In both cases I was talking about local models, deepseek-r1 32b parameter vs an equivalent that is uncensored from hugging face

I'm not talking about the cloud version at all. I'm talking about the 32b and 14b models vs ones people have "uncensored".

I was hoping someone knew of an "uncensored" version of deepseek that was good, that could run locally, because I haven't seen one.

I don't know what you mean by "removed".

I'm not talking about the speed I'm talking about the quality of output. I don't think you understand how these models are transformed into "uncensored models" but a lot of the time using abliteration messed then up.

Running an uncensored deepseek model that doesn't perform significantly worse than the regular deepseek models? I know how to dl and run models, I haven't seen an uncensored deepseek model that performs as well as the baseline deepseek model

I know how to download and run models, what I'm saying is, all the "uncensored" deepseek models are abliterated and perform worse

I haven't seen a way to do that that doesn't wreck the model

Every python library that does any real work is optimized to hell. The amount of power any ML training takes has nothing to do with python. You could write it all in assembly and I doubt you would gain more than 0.01% more efficiency and it would be way more difficult to develop and test new methods, and it would be way more error prone and likely waste more energy because of that.

SLA is also a mess

Same, I've got a modded 2080ti with 22gb of vram running deepseek 32b and it's great... But it's an old card, and with it being modded idk what the life expectancy is.

If AMD was smart they would release an upper-mid range card with like 40+ gb of vram. Doesn't event have to be their high end card, people wanting to do local/self serve AI stuff would swarm on those.

Great show, great first episode

The code is open source, people look at the code, I've dug through their code a fair bit. It wouldnt be quiet, and it would take major code rewrites, it would be pretty obvious

Accurate. And if you are being targeted by 5 eyes, your phone is probably fucked, one app vs another probably won't make a difference

Where is an uncensored version? Can you ask it about politics?

I was thinking the same. Next year it'll be measured in milliseconds

It's that advertising money... Ads should be heavily regulated and taxed

Agreed, and thanks for adding the background. Looks like someone abliterated deepseek already trying to make it "uncensored" but it sounds the process ruined the model.