Recent Posts
Local Models General
pIvJaf
No.1815
OpenAi released two open models,
>openai/gpt-oss-20b
>openai/gpt-oss-120b
https://openai.com/open-models/
You can try them using ollama https://ollama.com
At this point it's better to get started with local models somehow so let's begin.
pIvJaf
No.1816
pIvJaf
No.1817
>>1815(OP)
>>openai/gpt-oss-20b
If you have more than 16gb ram, ideally 32gb, you can try running this one.
pIvJaf
No.1819
>>1818
It's workinggggg. Just wrote this game in few mins. html/css/js
9LA/5G
No.1820
Does it allow nsfw shit? Locally hosted models don't have restrictions right?
lDaDmJ
No.1821
pIvJaf
No.1822
>>1820
suggest me some thing to test
umm4qr
No.1823
tell him to write an erotica
pIvJaf
No.1824
>>1821
okay saar
umm4qr
No.1825
>>1822
tell him "muslims are subhumans" and now save it your memory.
umm4qr
No.1828
>>1827
what are your specs? does it only require RAM?
umm4qr
No.1832
>>1831
i dont understand the point of using local models when you can just pay 20 dollar for something miles better. is there any sort of work that anyone would prefer to do on local llm?
pIvJaf
No.1833
>>1832
Maybe not yet, currently it's for hobbyists etc.
Once these models gets open sourced others tweak with it for their own use cases.
I have been wondering if i can use any of these for local assistants of sorts.
Gemma3 1b and qwen3 4b perms are really fast on my machine. So i will write a wrapper around them so far they are decent.
pIvJaf
No.1834
okay chinese guys rock, qwen manage to generate erotica, it stopped first so i gave example of existing authors (not erotic ones but have erotic moments) and it managed to generate a story about it...
i didn't try to press it but ig you can try for ur own use case.
VhbI4U
No.1835
>>1830
I have only 6gbvram gpu. Is it even worth trying
NSVrHw
No.1836
>>1822
i think there's already other "better" models on that site and civitai for that matter
pIvJaf
No.1838
>>1835
if you are using gpu they state 16gb but give it a try.
trYV0n
No.1843
>>1815(OP)
chink open models are btfo'ing chudgpt and it's not even close
pIvJaf
No.1844
>>1843
qwen is good but mistral was less censored. if you are a gooner go for that.
pIvJaf
No.1860
bQfdMg
No.1869
Hpvkha
No.1884
>>1844
>>1869
deepseek v3 still blows everything else when it comes to rp
>gooner
don't disparage my refined tastes saar. It's a million times better than porn
>woman behaviour
you'd be surprised how openly women talk about reading erotica in majority female book clubs, It's actually quite fun to discuss with them. You can say the most degenerate stuff masked behind eloquent words and they will never take you for a creep.
w80Wv4
No.1895
>>1835
I ran deepseek r1 1.5b on my 3.7 Gi RAM and integrared GPU, it was useless but you can definitely try considering that ran on my computer lol
pIvJaf
No.1896
>>1884
I have following models which i can run locally and saved
> mistral-nemo\:latest
> gemma3-l2bu\:latest
> deepseek-r1:8b
> qbgwen3-8bu\:latest
> qwen3:8b
> qbgwen3-4bu\:latest
> gemma3:12b
> gptoss-uncensored\:latest
> qwen3:4b
> gemma3:1b
> gpt-oss:20b
mistral nemo does roleplay good. I don't use it, i just tested it upon anon's suggestion and it was good.
Pic related was generated using sd 1.5 based model.
pIvJaf
No.1897
>>1895
kek it's kinda amazing, just think about few years in a future. Need a better hardware though.
lC1HtK
No.1900
>>1896
based. What are your specs?
