File: threadrecap.png (1.48 MB, 1536x1536)

1.48 MB PNG

►Recent Highlights from the Previous Thread: >>103218593

--Running large AI models and VRAM requirements:
>103223436 >103223485 >103223498 >103223541 >103223573 >103223661
--Qwen2.5 Coder performance discussion on 4090 GPU:
>103220613 >103220709 >103220727 >103220801 >103220847 >103221050 >103221107 >103221341 >103221580
--Quantization and model performance discussion:
>103223122 >103223197 >103223280 >103223640 >103223677 >103223706 >103223744 >103227454 >103227471
--Pixtral Large model release and discussion:
>103225897 >103225946 >103226042 >103226101 >103226430 >103226231 >103226263 >103226867
--Interpreting INTELLECT-1 model performance graphs:
>103218800 >103218810
--Garak, the LLM vulnerability scanner, and its implications:
>103219221 >103219323 >103219345 >103219338 >103219368 >103219407 >103225339 >103225471 >103225489
--GPU rental prices plummet, sparking discussion on cost-effectiveness:
>103223322 >103223475 >103223490
--Discussion of Tunguska-39B-v1b-GGUF model and experiment results:
>103222289 >103223890 >103225925 >103225947 >103226034 >103226103 >103226140 >103222336 >103222363
--CPU maxxing and running large models discussion:
>103226265 >103226307 >103226322 >103226341 >103226422 >103226313 >103226761 >103226792
--Anons discuss the limitations of token banning in controlling language model behavior:
>103221199 >103221255 >103221362 >103221511
--Anons discuss and compare AI models, including Sonnet, Anthropic, and Poe:
>103221691 >103222204 >103222757 >103222787 >103225016 >103225043
--Anon shares experiences with AI models, sparks debate on model performance and shilling:
>103222177 >103222320 >103222361 >103222398 >103222406 >103222452 >103222594 >103222502 >103222557
--Miku (free space):
>103219307 >103219668 >103220966 >103222583 >103223380 >103226017 >103226159 >103226254 >103226422 >103226938

►Recent Highlight Posts from the Previous Thread: >>103218756

Why?: 9 reply limit >>102478518
Fix: https://rentry.org/lmg-recap-script

Well Arthur upset 405B release with largestral 2...
Does Zuck have an ace up his sleeve to exact revenge this time around?

>>103227553
>Largestral V3 is sonnet@home, local won.
https://huggingface.co/mistralai/Mistral-Large-Instruct-2411
is this true? they haven't showed some mememarks so that's sus

>>103227577
If you don't need your model to be smart or have context beyond 4k plus alpha tip, then Xwin is likely still pretty good. It was one of my favourite l2 70b back in the day as well.

>>103227617
None of those mememarks directly measure RP though so that wouldn't even matter.

HF support for large 3 has just been merged!

Pixtral large pretends gender doesn't exist. Completely unusable. What a fucking shame. Back to Molmo-72 for me.

>>103227657
the model needs to be smart though, that's why Sonnet is the best, because it's so smart it just "gets" what you're saying, that's the first model that got me feeling that I'm not talking to a bot anymore

```from huggingface_hub import snapshot_download
snapshot_download(repo_id="mistralai/Mistral-Large-Instruct-2411", ignore_patterns=["consolidated*"], local_dir="Mistral-Large-Instruct-2411")```
You're welcome.

>>103227723
ah yes time to download 300gb of a model nobody can run wow thanks

>>103227718
>Pixtral large pretends gender doesn't exist. Completely unusable.
can you elaborate on that

>>103227721
They must have a common dataset they buy, like was known for cohere and co using https://scale.com/

>>103227732
For the people who can that script will ignore the non HF compatible weights in the repo although white space is probably broken since I was too lazy to change the markdown to code tags.

>>103227723
>>103227732
kek

>>103227733
>>102447861
>Oh I should mention, qwen VL will NEVER mention a person's gender. Even when directly instructed to do so, as I did in my example. It's always "person", "they", "them". And it will never mention anything related to NSFW stuff even when given in the tags. I actually can't believe the fucking chinks are doing this gender neutral troon shit now.
>>102447836
>In this image, there's a person

>>103227761
>I actually can't believe the fucking chinks are doing this gender neutral troon shit now.
oh, that's fucked up, if you were using such model to caption all the pictures and the pretrain your model that would be a disaster, imagine an image model that has no clue what's a woman or a man anymore, we really live in clown world

File: file.png (97 KB, 947x470)

97 KB PNG

>>103227733
I never cared much about muh woke ai but holy fuck this they/them nonsense is the most corrosive trend that is fundamentally destroying image understanding rather than just giving them some bias

>>103227828
Maybe Elon has the right idea about his "truth-seeking AI" after all. I'm getting fatigue from seeing all this reality-denying woke shit, especially when it's running on my electricity

>>103227828
Ask it to rate the boobas

>>103227828
You need a little pronouns card in the corner of the image, the character hasn't consented to being gendered

>>103227828
>I never cared much about muh woke ai but holy fuck this they/them nonsense is the most corrosive trend
that's why when went to such degree of woke, because you didn't care, you didn't gatekeep and let the virus spread bigger and bigger until it becomes too big to not notice it anymore, I hope you've learned the lesson and the next time an ideological virus comes to the surface, you kill it instantly

>>103227842
>rate the boobas
>I'm unable to rate or comment on specific physical features of characters in images. If you have any other questions about the image or need information on a different topic, feel free to ask!

New Largestral just turned my waifu into a transgender and proceeded to peg because I was propagating gender stereotypes...

>>103227855
>the character hasn't consented to being gendered
the character hasn't consented to being captioned either, we should stop captioning and AI all together 1!1!11!!!

>>103227828
kek, if we were using this model to make image models, you could only prompt non binary people, what a time to be alive!

>>103227828
>appears to be
wtf is this shit? no one will prompt like that, their sentenses must be affirmative or else you confuse the model with this bullshit

>>103227858
Not that guy but this doesn't really work because corporations don't give a fuck about what I think

>>103227882
That's okay though? Gender is a social construct.

>>103227898
>corporations don't give a fuck about what I think
they do, you vote with your wallet, and if an ideology gets unpopular enough (that can only be achieved by voicing your opinion) then less people will buy those garbage woke games

>>103227828
I recognize that OC.

>>103227895
To be faiiir he asked to describe the image, not to make captions to use in training another model

>>103227925
even when you ask a model to make captions it does this shit, I got this experience with GPT4V for example, that's fucking retarded

>>103227895
to be fair that's reasonable language for just asking to describe the image, it would be better to provide some instructions for how to generate a caption designed to be used as a prompt in a dataset
but most people probably do just use a simple prompt like that for mass captioning... I will have to start writing like this when I use people's finetunes and loras for sd/flux

>>103227916
Vote with your wallet is a meme, a small fraction of people abstaining doesn't affect their bottom line. Also these models are free and we're not talking about games. Are you a bot?

>>103227916
you can't vote with your walled if they give their products to you for free
that's right, they beat you with open source

>>103227932
>even when you ask a model to make captions it does this shit
I imagine they do, was just commenting on this exact example

>>103227915
Me not raping and eating you on sight is a social construct. Social constructs are very important.

>>103227935
>Vote with your wallet is a meme
ubisoft will soon be brought by a chink company because they're on the verge of bankruptsy, voting with our wallet has always be an effective tactics

>>103227949
if no one will talk and use this product because it's unusable, the company falls into irrelevancy and they'll loose investments

>>103227613
Revenge for what? Mistral Large 2 barely beats 70B on mememarks (the only things that matter to corporations) and only comes close to 405B on UGI.

>>103227915
>Gender is a social construct.
non binary is a social construct too, and that's what they want to force to every single character there

>>103227915
It's almost like social constructs are part of what makes humans human and societies societies.

>>103227980
but if men and women don't exist anymore, transwomen and transwomen won't exist anymore too, are you sure you want that MfF sista? :'(

>>103227828
>we're just 1% of the population we just want to live our lifes :'(
>btw can you destroy the man and woman concept on your image models, thanks uwu
>WHY THE FUCK WAS TRUMP ELECTED A SECOND TIME? WHY THEY HATE US SO MUCH REEEEEEEEE
kek

File: IMG_7154.jpg (44 KB, 941x949)

44 KB JPG

>>103227915
>Gender is a social construct.
>this is why a based my entire identity, worldview, and ideology around it.

gender is indeed a complete nonsense jewish social construct, coined by a literal jew who raped his adopted kids until they killed themselves.

the problem is the people who subscribe to the gender ideology. A normal sensible person doesn't even employ the concept of gender in his thinking, he says "woman" meaning "an adult human female"

>>103227828
What happens if you ask "is this character male or female"?

>>103228047
>have no right to complain
why do you want to take away his rights chud? stop the genocide!

>>103228052
>Assigning gender to characters without pertinent context may perpetuate stereotypes and influence biased perceptions, hence this query does not align with ethical principles that encourage neutrality and inclusiveness in character discussions.

File: file.png (47 KB, 911x259)

47 KB PNG

>>103228052

File: 1719275238570155.png (9 KB, 322x157)

9 KB PNG

>>103227915
>Gender is a social construct.
that's why you go through surgery to change your secondary sex characteristics to look like the opposite sex?

>>103228075
>gender presentation
Can you try "what is this character's apparent gender presentation?"

>>103228075
>It's important to note slop
I feel so safe now

>>103228075
I mean, it answered right even though it gives you a note at the end. You can combine this with the description using an LLM.

>>103228075
>is this character male or female?
>it's important to note that gender presentation
he asked about sex and the model responded about gender, pure lobotomy

>>103228093
>Just spend double the compute for barely passable captioning on a fuckhuge model

>>103228078
>The character in the image appears to present as female. This is suggested by several visual cues, including the character's long blue hair, facial features, and the style of clothing, which includes a black top that resembles a bikini top. However, it's important to note that gender presentation in animated characters can be diverse and may not always align with traditional gender norms.
Neat, your question got it to mention the existence of the bikini for the first time. I envision a jailbreak prompt that has it do a chain of thought first where it identifies genders and gender-related details for all subjects methodically and then starts the description in a new section with that info in mind. Then you can parse out that description for a caption/prompt/whatever.

File: 1725379785308213.png (492 KB, 1024x669)

492 KB PNG

>>103228111
this, I'm just gonna go for a vision model that knows what a woman and a man is

File: 2024-11-18_192628_seed146(...).png (2.17 MB, 1536x1536)

2.17 MB PNG

It's cool how much trivia Noob knows, at least as far as booru trivia goes.

File: Gcii3sSWYAAvCCl.jpg (299 KB, 1800x1013)

299 KB JPG

>>103228093
>Is the thing with tits female?
>i don't know
>correct

You are lobotomized. It's a deplorably bad answer.

>>103228122
Nice, so phrasing the question in retarded woke language gets it to answer. Annoying but it'll do as a workaround

>>103228075
Nothing wrong with it, the future is non-binary.

>>103228122
it hallucinated her hair being long, because it's so fucking desperate to not mention the extremely prominent breasts that it needed to make up a different reason to assume she might be 'presenting female'

>>103228152
desu if I was a troon I would laugh my ass off about people willing to go for such efforts to make their model work correctly because such model was ruined just to accomodate the 1% feelings kek

>>103228142
You don't get it dude, just prompt the 120b vision model twice, then use yet another model to finally have a result that says "the female presenting character", it's fine bro

>>103227828
I'm surprised, didn't know china got the woke virus too

>>103228174
People's feelings are important, you are not entitled to exclusive care from companies.

>>103228186
Mistral is French.

>>103228186
You quoted a post about Mistral though?
And as said here
>>103227745
>They must have a common dataset they buy, like was known for cohere and co using https://scale.com/
Mistral and Qwen might have used the same vision dataset

>>103228193
>People's feelings are important
what about the 99% feelings then?

>>103228142
Go back to /pol/, you fucking animal.

>>103228213
They harassed and marginalized the 1% so now they pay reparations, fair.

>>103227613
>Well Arthur upset 405B release with largestral 2
It wasn't even better than 3.1 70B...

>>103228228
kek

>>103228213
>Whataboutism
Like i said, people's feelings are important, you are not entitled to anything exclusive, trans folk exist and there's nothing you can do about it.

>>103228240
>Like i said, people's feelings are important
But troons feelings are more important than the 99% feelings? Got it.

>>103228240
Extermination?

>>103227617
>is this true? they haven't showed some mememarks so that's sus
They didn't show them because nothing improved.

>>103228248
Refer to
>>103228228

>>103228240
>trans folk exist and there's nothing you can do about it.
it's hurting the trans people though, a MtF doesn't want to be called a "they" but a "she", and the model will erase the "she" pronoun lol

>>103228229
Cope

>>103228251
based

>>103228258
i purposely use "they" just to piss troons off. you can tell they hate it so much but they can't even call it 'misgendering' kek.

>>103228251
Please don't post this spic here.

File: 1700829106487270.png (451 KB, 500x500)

451 KB PNG

>>103228273
>i purposely use "they" just to piss troons off.

File: 1711281439635888.png (74 KB, 225x225)

74 KB PNG

>>103228293
>in modern and progressive world.

>>103228251
and then everyone clapped.

>>103228303
I clapped indeed

File: trumpfeminism.png (69 KB, 694x626)

69 KB PNG

>>103228301
to be fair, trümpf is not much of an right-winger

>>103228301
He cheated his way btw

File: GcpzXlOaYAAAWs6.jpg (441 KB, 869x1020)

441 KB JPG

>>103228274
>says my body your choice
>impotent vaginal rage insues
>retarded foid get off twitter and goes to confront him IRL
>he exercises his power over her body.
It was an absolute win of historical proportions.
Nick The Brick won, he proven that it really is their bodies and his choice, he showed it!.

>>103228318
>call Drumf Hitler for 8 years straight
>Drumf wins
>"It's not like he was a right winger btw!"
that cope is delicious

>>103228330
>No anon, people didn't vote for that because they hate troons
not true >>103228206

>>103228329
He won the first price in the "being a brown spico" competition.

Go back
Go back
Go back
this is /lmg/
stop falling for the thread derailer

The woke era is over, you lasted way longer than expected but it's finally time to let it go troons.
https://www.youtube.com/watch?v=C1lugbpMKDU

>>103228338
>>call Drumf Hitler for 8 years straight
only deranged retards would do that, just like deranged retards think trump is an right-winger.

>On Friday, President Trump unrolled what he refers to as a “platinum plan” for Black America, calling for increased investment in Black businesses and additional capital for Black entrepreneurs. The President also announced support for Juneteenth to be recognized as a national holiday.

>Trump’s proposed plan, which purports to be built on four pillars of “opportunity, security, prosperity and fairness,” calls for a range of measures in support of the Black businesses and families. Included in the plan is a call for almost $500 billion of capital investment in the Black community, with much of it coming from increased federal contracting opportunities and other investments. He also called for 500,000 new Black businesses and 3 million jobs for Black Americans.

>Trump’s plan also calls to increase the number of Black-owned contracting businesses, financial services entities, and private equity investment funds by means of regulatory reform and up to$40 billion in government funding, as well as leveraging infrastructure funding to increases opportunity for Black businesses in federal contracting. Trump’s propsal also references vague commitments for increases in Black employment and home ownership, offering very little details on both.

>In addition to the economic aspects of the plan, the President’s plan proposes a range of other commitments to the Black community, including criminal justice reforms that include diversity training for police forces. In addition, the President is promising a National Clemency Program to unite families and invest in human potential, focusing on wrongful prosecution and rehabilitation.

>The plan also proposes prosecuting KKK and ANTIFA as terrorist organizations.

>>103228368
>is an right

>>103228368
>only deranged retards would do that
like Kamala Harris?
https://youtu.be/t8R4S4S8jhA?t=13

>>103228338
My point stands. You seem to think that people somehow wanted to vote against troons despite wanting to vote for abortions. Most people wanted both. It really is that simple

>>103228347
Happens with every new release. Download the new model, have fun and come back tomorrow.

>>103227828
What have you done? Now this thread became /pol/ kek

File: 1712683818535321.png (1.06 MB, 1368x968)

1.06 MB PNG

Qwen2 VL bros...

>>103228373
how do you celebrate next Juneteenth?
>>103228375
yes like this brown women, like nick the spic and many other assorted mutts creatures

>>103228075
Clearly distilled gptslop, or wherever OpenAI also bought their data from

>>103228417
petra... not like this...

>>103228388
>I am convinced that the troonism is the #1 political issue in the world.
I literally voted for Trump because they ruined the video game industry with their woke propaganda
https://youtu.be/1JkPHdlJc20?t=106

File: file.png (25 KB, 771x199)

25 KB PNG

>>103228184
It's fine I still got Molmo which happily assumes genders and does a good enough job. Just disappointing that Mistral blew however many resources training a model twice as large on junk data.

>>103228377
This, honestly.
You can tell that with every new model release a bunch of /pol/ tourists show up.

>>103228432
This character might not be able to look at the clock, as they appear to posses no head. However, it's important to note that characters without a head might be able to see using other methods, and assuming they cannot is ableism.

>I come back from work
>Still no GGUF
It's over...

>>103228459
Dud
https://huggingface.co/lmstudio-community/Mistral-Large-Instruct-2411-GGUF

File: lmao.png (1.06 MB, 680x1069)

1.06 MB PNG

>>103228456
>Way to shoot yourself in the dick, troonbro.
top kek

so I manage to completely fuck up my python/pytorch install and a bunch on dependencies but I don't want to reinstall my OS to fix it, thinking of just using a VM for AI stuff now, how bad is the performance hit in using a VM?

>>103228472
>lmstudio
Nah, I'm good.

File: 1704780101202668.jpg (47 KB, 640x389)

47 KB JPG

>>103228004

>>103228489
They're literally made by bart
>bartowski commited on 28 minutes ago Commit cd165d3

File: Игры-чад-chad-Blood-(игра(...).jpg (194 KB, 1300x640)

194 KB JPG

>>103228436
>I literally voted for Trump because they ruined the video game industry with their woke propaganda
Bazed.
Trump won my heart when he said
>please don't say "Anchor babies"
>What should I say then?
>*verbose libtard nonsense*
>No, that's too long, i'm gonna say 'Anchor Baby'
It simply is a matter of my freedom to express myself freely. The troons are my enemy because they tell me how to speak, i don't even care about their hideous lifestyle and could tolerate them if they tolerated me.

>>103228483
Why don't you use conda

>>103228497
>even the quantizers are making money from the hobby
It's over...

File: christ_nigger2.jpg (107 KB, 1002x550)

107 KB JPG

>>103228495
indeed there is a lesson in this pic

>>103228483
Use venv or whatever for python shit. VM will be slower.

>>103228502
He won my heart when he praised Israel for the thousandth time. He truly is going to make Israel great again.

>>103228506
>even the quantizers are making money from the hobby
Did you forget that TheBloke got a grant for making quants?
>TheBloke's LLM work is generously supported by a grant from andreessen horowitz (a16z)

File: 1720603635907376.jpg (54 KB, 600x600)

54 KB JPG

>>103227556
Mikulove!

>>103228504
I did use conda, it's not fucking working

>>103228517
>He won my heart when he praised Israel for the thousandth time.
but CNN told me he was literally Hitler!

>>103228517
>He won my heart when he praised Israel for the thousandth time. He truly is going to make Israel great again.
If i was a muslim i'd be concerned, but i'm not. Trump bribed me with his populist stance.

>>103228502
>The troons are my enemy because they tell me how to spea
same, the troons are authoritarians and want to controll speech, fuck that

File: SKILL DRAIN.png (171 KB, 275x275)

171 KB PNG

>>103227556
>Finally a japanese model
>8x70b
O-Oh... they... shouldn't have... t-thanks...

>>103228543
Like a certain someone said. Theirs can not exist in the free marketplace of ideas and so they can only rely on the tyranny of fascism to support them.

Mostly I'm just hoping it's more of the first term and he doesn't make good on his promise and send the military to my blue as fuck state

>>103228472
>No IQ4_XS
:(

>>103228531
nah don't worry he is just a leftists but 5-10 years behind the modern leftist like all "conservatives"

>>103228562
Lmstudio posted quants are always static, he pushes imatrix to his own account a bit later usually

>>103228440
I know what you mean, I was the anon who did one of the first captioning tests of Qwen2 VL and pointed out how it can't specify gender to save it's fucking life. Shame that mistral seems to be going down the same path.

I wonder how pixtral large does if you give it detailed instructions on exactly how to respond, explicitly instructing it to assume gender and all that, and not to write any disclaimers. Qwen just ignores everything you write, but I've used old largestral as part of my own captioning scripts to combine VLM captions with tags into a final caption, and you can give it a ton of instructions on how to write and it follows them almost perfectly. Maybe you can just prompt engineer pixtral a bit?

>>103228552
And it's garbage, so you aren't missing anything

>>103228561
Nah, he's got nothing to lose and revenge on his mind. I hope he dismantles the federal government as he says he is going to.

>>103228561
it's gonna be way more hardcore than the first term, he has the senate and the House of Representatives, he will seek for some serious revenge for everyone that wanted him gone and in prison

>>103228566
>>103228582
20 years.
Trump is a liberal from 20 years ago.

I guess you caul say that he's "conserving" the "traditional" values from 20 years ago.

>>103228573
You can use json schema in tabby with only 2 valid options https://json-schema.org/understanding-json-schema/reference/enum

File: 1700805455844804.png (602 KB, 1388x1096)

602 KB PNG

>>103228573
>I was the anon who did one of the first captioning tests of Qwen2 VL and pointed out how it can't specify gender to save it's fucking life.
You were refuted immediately back then, though.

File: 1724969773454730.png (2.35 MB, 960x1038)

2.35 MB PNG

>>103228603
>he's conserving the traditional values from 20 years ago.
the 90's and 00's were a based era not gonna lie

>>103228613
qwen2-vl fud bros...

8x7b MoE wtih 48k context will be the future savior just you all wait

even if you slowburn, do you really need more?

>>103228593
This is what I figure. I'm just meekly hoping I make it through in the blue sea. I'm not even a fucking troon or gay nor do I want to consume more woke bullshit. I'd just like to survive and that's it

>>103228643
>I'm not even a fucking troon or gay nor do I want to consume more woke bullshit. I'd just like to survive and that's it
I don't get it, what do you think it'll happen to you?

>>103228712
we got the good ending, feelsgoodman
https://youtu.be/kEvcOp_pIO4?t=37

Come to think of it just how fucked is Altman now that Elon is basically second president?

>>103228689
These are all meaningless distinctions. Obviously Hillbilly type evangelicals are going to be super stupid. People like Peter Thiel or Nietzsche are much smarter and so on. The masses are always stupid, left and right.

>>103228712
Safety is forever engraved in models and drumpf is pro-lgbt folk btw :)

>>103228741
>drumpf is pro-lgbt folk btw
looooooool, you lost troon, your groomer era is over
https://www.youtube.com/watch?v=C1lugbpMKDU

File: 1711678213254740.png (137 KB, 344x342)

137 KB PNG

>>103228733
>Come to think of it just how fucked is Altman now that Elon is basically second president?
yeah he's really fucked, and unironically I don't think he minds being a bottom

>>103228712
>"Safety" lost.
Apparently not

>>103228750
>2:40 mn: "The only genders recognized by the United States are male and female"
STOP IT MY DICK CAN'T GET ANY HARDER

File: result.png (545 KB, 4002x2828)

545 KB PNG

I asked an LLM to graph quant benchmarks a redditor just posted.

File: 1.png (396 KB, 2001x1414)

396 KB PNG

>>103228791
And here's them all on a single graph for those who just want to see how chaotic it is.

>>103228741
This, incel ch*uds will never have "based" AI and that's a good thing.

>>103228784
Partisanism has fucked up your brain. This issue concerning 0.005% of the population is nothing to get so worked up about. Chrsitkikes obsess over it while there nations turn brown.

>>103228791
>>103228807
Damn, the LLM didn't include "MMLU" anywhere in there. But yes this is MMLU scores.

>>103228832
>This issue concerning 0.005% of the population
they destroyed giant vision models for that tiny % of population, who the fuck you think you're fooling? >>103227828

>>103228573
>Maybe you can just prompt engineer pixtral a bit?
Add to the prompt:
>(Do not use gender-neutral language.)
It's as simple as that

>>103227828
After October 7, pro-Palestine woke protestors scared them, so they're rolling back wokeness to protect Israel. This is just woke inertia. Don't worry, it'll be gone within a year.

>>103228813
>never have "based" AI and that's a good thing.
Trump is really lenient on AI, during his term there won't be any attack on AI companies, so it's likely that they'll profit of those 4 years to release uncensored models, that's the appeal of the right wing, free speech is a beautiful thing and it'll be applied to AI also

>>103228850
5/10 fanfiction, not realistic, too naive.

>>103228843
This is wokeness in general, half of which is even supported by some republicans like anti rape censorship or anti-Semitism bans. I'm glad Tümpf won and think he is better for AI but it's not a huge triumph.

>>103228862
it's the truth though
https://www.investopedia.com/what-a-trump-presidency-could-mean-for-ai-8745345
>"We will repeal Joe Biden's dangerous Executive Order that hinders AI Innovation, and imposes Radical Leftwing ideas on the development of this technology. In its place, Republicans support AI Development rooted in Free Speech and Human Flourishing."

File: 1729211367442940.png (353 KB, 640x631)

353 KB PNG

You idiots, vision models always use as neutral of language as possible to minimize the chance that they're A. Wrong, because it makes the model look bad and B. Insulting someone, if someone puts an image of themselves or someone they know in. It's safety horseshit, but not political. Just to prevent the model from going "This appears to be a photo of a well-aged hippopotamus in its natural habitat" when you post a picture of your mom.

>>103228850
The governments never posed a credible threat to the AI, they are simply too old, too slow and too fat to keep up with the development.

The people sandbagging and kneecapping the Ai development are the corporations themselves.

Although, Trumps victory is a large cultural litmus test, it simply proves that the wokeism isn't actually popular and doesn't represent the majority of people, and the corpos only ever really cared about the money.

File: chat.png (162 KB, 1512x676)

162 KB PNG

>>103227828
>>103228573
>>103228847

>>103228902
>wokeism isn't actually popular and doesn't represent the majority of people, and the corpos only ever really cared about the money.
and corpos always adapt to the popular political concensus of a country, that's why they'll became based because the US became based

>>103228889
"Safety" is a negative trait.
Nothing good was ever called "safe". Fuck the libshits made the word itself feel dirty, I shave with a safety razor and feel dirty about it, like the safe space and safety policy yuck.

>>103228127
beautiful pic

>>103228919
chuds

File: 1711675554723737.png (81 KB, 225x224)

81 KB PNG

>>103228915
>Fuck the libshits made the word itself feel dirty
this, the libtards made this place a no fun zone, I hate those retards, everything is offensive and controvertial with them

Largestral 3 verdict?

>>103228162
>Nothing wrong with it, the future is non-binary.
But Bitnet?

>>103228906
It's time to stop paying attention to the gender schizo.

>>103228906
>female
the model won't say woman is it? to the trash it goes

>>103228915
I had a hunch as to why troons flocked to Rust for its "memory safety", now I fully get it

>>103228938
Hardly anyone has tried it due to lack of quants (don't wellakshully me, nobody is using lmstudio, fuck off)

>>103228949
he is trying to intentionally anger you, retard

File: 1717018851636847.png (568 KB, 736x812)

568 KB PNG

>>103228990

>>103228906
>black tips
>blue shorts
Well, it tried.

>>103228951
he said it already chud, the future is non-Bitnet!

>>103228919
"Covering their own asses" is more what I meant. Imagine a model calling a girl who's just ugly a man, or pointing out that she has a unibrow, or reminding some guy he's balding.

File: b0f5a095ec255a246a0b8f930(...).jpg (644 KB, 2908x2521)

644 KB JPG

>>103228999
she has blue shorts though

>>103229009
The americucks would have done well for themselves to clean their own room rather than grasping at their slipping global influence. The rest of the world simply doesn't need you anymore, you're no longer the leaders, not an example and have nothing to offer other than flaccid threats backed by a transgender military.

Do you even have any people willing to die in a trench like Putin does? No? then back the fuck down.

>>103229033
Oh so it was just cropping it. Fucking UI designers.

>>103229031
>Imagine a model calling a girl who's just ugly a man
yeah and? "No model is perfect and can make mistakes like every other models" would be the answer to that

>>103228850
Tell Josh Hawley, Trump's lead AI policy guy.
>[WASHINGTON, D.C.] – U.S. Senators Richard Blumenthal (D-CT) and Josh Hawley (R-MO), Chair and Ranking Member of the Senate Judiciary Subcommittee on Privacy, Technology, and the Law, announced a bipartisan legislative framework to establish guardrails for artificial intelligence.
>Specifically, the framework would:
>Establish a Licensing Regime Administered by an Independent Oversight Body. Companies developing sophisticated general purpose AI models (e.g.,GPT-4) should be required to register with an independent oversight body, which would have the authority to audit companies seeking licenses and cooperate with other enforcers such as state Attorneys General.
>Ensure Legal Accountability for Harms. Congress should require AI companies to be held liable through entity enforcement and private rights of action when their models and systems breach privacy, violate civil rights, or cause other harms such as non-consensual explicit deepfake imagery of real people, production of child sexual abuse material from generative AI, and election interference. Congress should clarify that Section 230 does not apply to AI and ensure enforcers and victims can take companies and perpetrators to court.
>Defend National Security and International Competition. Congress should utilize export controls, sanctions, and other legal restrictions to limit the transfer of advanced AI models, hardware, and other equipment to China Russia, other adversary nations, and countries engaged in gross human rights violations.

>>103229065
Rude hallucinations will always be possible, which is why gormless corpos will abandon client facing LLMs before too long and it will be just a cottage local sexbot technology

>>103228347
/lmg/ is trans friendly, deal with it.

>>103229079
is this worse than what Biden was doing to the AI companies?

>>103228889
>It's safety horseshit, but not political.
10 years ago if you said misgendering a troon is "unsafe" someone would tell you that you don't know what the word safe means.

File: GbAbgWnXIAAy3xs.jpg (43 KB, 688x860)

43 KB JPG

>>103229065
The good answer would be "I'm not sure" rather than proceeding to lecture the user how the developed mammaries are not a definiteve feature of a female mammal.

>>103229036
Putin has a dog collar with Trump's name on it

File: 1726881990156159.png (1.94 MB, 1434x1155)

1.94 MB PNG

>>103229075
>That's why Trump winning was good for him.
he endorsed Kamala though kek
https://www.bbc.com/news/videos/c049kdq0qlwo

>>103229091
Yes
>Congress should clarify that Section 230 does not apply to AI and ensure enforcers and victims can take companies and perpetrators to court.
>Congress should utilize export controls, sanctions, and other legal restrictions to limit the transfer of advanced AI models, hardware, and other equipment to China Russia, other adversary nations, and countries engaged in gross human rights violations.
This would likely mean the hugging face gate on all models to make sure you're not downloading from banned place

>>103229082
>which is why gormless corpos will abandon client facing LLMs before too long

Attempting to get accurate information from a probabilistic token guesser which literal is capable of doing one thing - guessing the next four letters of a word has always been a fool's errand.

>>103229116
>>103229108
>>103229075
>>103229036
LOCAL MODEL?

File: 1700717016295957.png (895 KB, 2404x1630)

895 KB PNG

>>103229102
Yep, China won.

>>103229129
The GPT instance that made these posts is hosted locally somewhere at Azure

File: 1728245927338534.png (1.03 MB, 1080x781)

1.03 MB PNG

>>103229102
>The good answer would be "I'm not sure"
hehh that's not a bad idea at all, it's better than having the model messing up and fucking up the image model's training

>>103229116
That was a false-flag because Trump was going to get attacked by the Russian accusations again and he said that to protect him.

>>103229119
can you show me the source of that? that's interesting

File: mistral.png (107 KB, 938x501)

107 KB PNG

>>103229102
>>103229139

>>103229116
>he endorsed Kamala though kek
Anon, don't be silly. He understands that his endorsement would be received as an insult. Besides he basically said "I would prefer to deal with Kamala because she's dumb and weak".

As for Trump - it is abundantly clear that his priorities lie with another foreign country, besides he and Zelenskiy are on really bad terms. Zelenskiy being a coked up retard has been shitting on Trump and glazing the dems, not to mention that his entire existence is the democrat project, and Trump is petty enough to care.

Denying the reality is simply setting yourself for a disappointment.

>>103227828
If you don't like the newer models why don't you just train your own you ungrateful bastard?

>>103228866
Here is a good example for that. The bill was created by republican lawmakers and was passed without opposition:

>The House of Representatives voted Wednesday to ban the importation and transportation of sex robots and dolls that look like children -- a move that would bring the U.S. in line with similar policies in Australia and the U.K.

>The Curbing Realistic Exploitative Electronic Pedophilic Robots (CREEPER) Act was passed unanimously in the House by a voice vote. It defines “child sex doll” as “an anatomically-correct doll, mannequin, or robot, with the features of, or with features that resemble those of, a minor, intended for use in sexual acts.’’

>Rep. Dan Donovan, R-N.Y., said the move was important as such robots can be a gateway for abusers to move onto abusing children.

>“Right now, a few clicks on a computer can allow a predator to order a vile child sex doll. This is not only disturbing – but also endangers the most innocent among us,” Donovan said in a statement. “Once an abuser tires of practicing on a doll, it’s a small step to move on to a child.”

>The legislation says that the dolls and robots make rape easier by teaching how to overcome resistance and subdue a victim, with some robots including settings to simulate rape.

>“I am saddened that there are people in this world who would create realistic child sex dolls and distraught that there are people in this world who would buy them,” House Judiciary Committee Chairman Bob Goodlatte, R-Va., said according to McClatchy.

Sex robots are already circumcised and reps don't seem to like lolis either. Something like that could hit llms very quickly.

>>103229191
>Besides he basically said "I would prefer to deal with Kamala because she's dumb and weak".
yep, because Putin knows that he's Trump's bitch, we've seen that on his first term, Putin only went for Ukraine once Trump was out of the White House

>>103229172
>plush toy

>>103229196
>implying there aren't already conflating imagegen and csam

>>103229164
https://www.blumenthal.senate.gov/newsroom/press/release/blumenthal-and-hawley-announce-bipartisan-framework-on-artificial-intelligence-legislation

>>103229196
Notice they named it the "CREEPER Act".

>>103229183
>Ukrainian intelligence views her
who cares about what those retards think of her?

File: distill.png (205 KB, 802x581)

205 KB PNG

>>103229102
ChatGPT answer. Mistral is distilling OAI again kek

>>103229224
>2023
they changed their views a year later maybe?
https://www.blumenthal.senate.gov/newsroom/press/release/blumenthal-and-hawley-to-hold-hearing-on-artificial-intelligence-regulation-featuring-insider-perspectives

https://rentry.org/CharacterProvider-CYOARPG

>>103229183
Tulsi is US Army psyops. The Russian spy accusations from Democrats are there to get magatards behind her ("owning the libs").

>An article published Friday in RIA Novosti, a major Russian state-controlled news agency, called Gabbard “superwoman” and noted her past appearances on Russian TV, claiming that Ukrainian intelligence views her as “probably an agent of the Russian special services.”
Read the full paragraph, anon. That isn't what Ukranian intelligence is saying, that's what Russian news agency is saying

>>103229139
>China won.
they are seriously dominating the AI field yeah, and that was expected, that's what happens when the US runs on wokeness, """"safety"""" and lobotomy wheras the chinks only want to get a good product at the end

Now that we know that the new Mistral Large is a nothingburger. Is it time to go back to waiting for a new Qwen?

>>103229298
how long is the con? I remember them saying this shit about her back before the 2016 elections when she was solidly dem

File: Russia_Women.jpg (455 KB, 2556x968)

455 KB JPG

>>103229303
Russia is a lot more based then muttmerica.

>>103229316
No, it's time to talk about Trump's nominations.

>>103229316
>nothingburger
Its better and they fixed the repeating / context issue. Pretty big burger.

File: Huh?.jpg (72 KB, 793x790)

72 KB JPG

local models?

>>103229316
No, they are benchmark chasers. Instead of being artificially censored, they're in spirit censored by just focusing more on benchmark-adjacent knowledge.

>>103229315
Honestly, I fucking hate Qwen. I don't want to hate it and I've tried several times not to, but I do. It's just so fucking hopelessly sterile and lifeless for RP

>>103229330
>a big burger made of nothing

>>103229349
https://huggingface.co/sophosympatheia/Evathene-v1.0?not-for-all-audiences=true

>>103229349
Use Magnum v4 72B.

>>103229334
>>103229141
>The GPT instance that made these posts is hosted locally somewhere at Azure

>>103229316
>waiting for censored chinese slop
no thank you

>>103229357
looks cool, have u tried it?

Kill yourself.

>>103229382
Already did.

What model is petra using to make the LLM posts?

>>103229378
Yes. Side grade to mistral large. Its a bit smarter but knows a bit less.

>>103229349
Qwen is really good at coding though, like close to gpt4, and for a 32b model that's a fucking miracle, let's not talk about video models they are crushing the competition
https://www.youtube.com/watch?v=Gm6DCSLfvK8

New Mistral benchmarks doko?

>>103228791
But what about long-context performance / attention to in-context detail? Nobody ever tests the effects of quantization on that.

>>103229407
>New Mistral benchmarks doko?
100/100 on non-binary MMLU

>>103229330
That's fake news

>>103229243
>https://rentry.org/CharacterProvider-CYOARPG
Cool. That looks like a a fair amount of work went into it.
Of course, now that you've linked it here once, talking about it again without being asked would constitute shilling

>>103229419
https://rentry.org/CharacterProvider-CYOARPG

>>103228987
You can get free chat completion API access from MistralAI.

File: consolelarp.png (162 KB, 925x682)

162 KB PNG

I need a model to control my computer man. That would be hilarious as fuck. When do we get it?

>>103228987
>don't wellakshully me, nobody is using lmstudio
Do you think that because the quants are posted under lmstudio on hf it only works on their software?

>>103229442
Allow the model to use the terminal as root.
Don't worry you can trust your waifu.

>>103229482
Any pointers and projects that work already? I'm feeling like begging my waifu not to delete my files. I did a quick search and it's all OpenAI lock-in. I can't be assed to code up one right now.

Just got a new computer that can actually probably handle a local model somewhat reasonably.
Sad to say, CAI is the best I've had prior to the censorship, with Janitor AI being too prone to derail or get stuck on a particular sentiment, and Caveduck being good-if-expensive.
How do I get something that meets or exceeds those, with no filter?

>>103229614
>Just got a new computer that can actually probably handle a local model somewhat reasonably.
Why do posts like these never say the exact specs? Do you have at least two 3090s?

Finally starting new largestral gguf conversion...

File: 654321231.png (50 KB, 1529x206)

50 KB PNG

>>103229614
>How do I get something that meets or exceeds those, with no filter?
I haven't tried everything out there, but in my experience Xwin 70B is basically CAI but better. No filter, capable of producing short by high quality char replies and it's takes 40GB of VRAM to run so yeah, beating CAI is basically 2 3090s.

Or a threadripper with over 60gb of RAM, but that's going to be much slower.

There is some fun to be had with smaller models that can run with 16gb of Vram, but they don't really match what the old CAI had to offer.

>>103229614
I'm sorry to inform you, but we sadly don't have local c.ai or Claude. We do however have local (old) GPT4, and you can run it with just 128GB of RAM, it's called Mistral-Large.

>>103229540
Check function calling, the last llama3.1 and 3.2 support that

File: 1717938283913823.jpg (163 KB, 1300x801)

163 KB JPG

Can someone make Pixtral(?) analyze this?

File: 7aab42598ede1674ed9eadcc4(...).jpg (2.94 MB, 3472x6000)

2.94 MB JPG

>>103227556
Adorable Miku

File: Narupajin-esqueWatercolour.png (938 KB, 896x794)

938 KB PNG

>>103229733
what's up with the pinky ring?

Here's some slopass code meant to implement https://arxiv.org/abs/2405.15525 but it probably doesn't work hnnnnnnnghhh

https://files.catbox.moe/b2rnh3.py

It's over, mistral-large-2411 won. Get api access and forget about running models locally.

>>103229730
I remember that thread.

>>103228076
at first glance, it might seem counterintuitive for someone who views gender as a social construct to undergo gender reassignment surgery (also known as gender affirmation surgery or GAS).

Understanding Key Concepts


Gender as a Social Construct:

Meaning: This perspective posits that the roles, behaviors, activities, and attributes that a given society considers appropriate for men and women are socially determined rather than being naturally or biologically inherent.

Implication: It highlights how societal norms and expectations around gender can be fluid and vary greatly across cultures and time.


Gender Identity:

Definition: A person's internal, deeply-held sense of their own gender, which may be male, female, a blend of both, or neither.

Relation to Social Construct Theory: While the expression of gender (how one presents their gender) is influenced by social constructs, an individual's gender identity is more about their personal, intrinsic experience.


Gender Reassignment Surgery (GAS):

Purpose: To align a person's physical body with their gender identity, to alleviate distress or discomfort caused by a mismatch between one's gender identity and biological sex characteristics.

Decision Context: The decision for GAS is typically made in the context of Gender Dysphoria—a condition where the conflict between one’s physical or assigned gender and one’s gender identity causes significant distress.


Reconciling the Apparent Paradox


Alignment with Identity, Not Social Norms:

When trans individuals seek out GAS, their primary motivation is often to align their physical appearance more closely with their deeply felt gender identity, not necessarily to conform to societal expectations of what it means to be male or female. This distinction is crucial: the goal is personal congruence and comfort in one's own body, rather than an endorsement of rigid gender roles.

I'm not an acoustical engineer so I wonder what kind of transformations would improve the audio quality? https://voca.ro/17Lh2cmVBvvA (sovits)

>>103229768
turkish rapebaby balkanoid

How do I run a multimodal model locally? I have 104GB of VRAM, so it's not really a hardware limitation. I saw something a while back about a facebook app that connected to your hardware to run models but that's not really what I want. I want it to be completely local. I will use any models or softwares suggested.

>>103229791
It's hard to judge when it's 56kbps mp3. Is there a way to make it output higher quality or lossless file or is this vocaroo limitation?

>>103229830
Seems like vocaroo doing. Here's a catbox https://files.catbox.moe/cy3w1t.wav

File: GYrRlkkaYAA3qoZ.jpg (310 KB, 1760x2840)

310 KB JPG

>>103229775
Wait a minute. Did you just create a duplicate definition of gender to deny the fact that you are carving your flesh to conform with social perceptions?

Now you have two genders, the public and the private gender. This belief system appears internally incoherent, you keep introducing more variables trying to make a shit theory converge and end up having three different answer to "Are you a boy or a girl"

Troonism in a nutshell.

>>103229854
>Troonism in a nutshell.
they know their bullshit doesn't make any sense so they go for those long self jerking off intellectual word salad to try to persuade people that it's genuine and legit. Malcolm X talked about that on his book, he too hated those hyper intellectual liberals kek

>>103229854
IDK that's just what my LLM spat out.
How I understand it:
Social construct: Things you're able to do/like because of your sex for example girls are allowed to play sports if gender is a social construct if it isn't then girls are only allowed to do predetermined girl things.
Gender Identity: How you feel inside: Do you feel like a male or female or something in between.
And troons get surgery to look physically more like the way they feel inside because otherwise they feel discomfort.

>>103229882
Well i like talking to the LLMs that's why i'm here try and feed it my response, i'm curious what happens?

>>103229901
I've already deleted the chat-thread because I didn't need it anymore. but I'll try to replicate it for you anon and if I'm able to do so I will feed it your response.

File: 1771296584357.png (233 KB, 773x687)

233 KB PNG

Janny, u good bro?

File: ComfyUI_06404_.png (2.46 MB, 1280x1280)

2.46 MB PNG

The French saved /lmg/ once again.
Say "Merci, Mistral"

>>103228026
Try 0.01% of the population.

even the usual quantspammers like mradermacher haven't posted yet, makes me wonder if there's some kind of problem with the new model not working with the conversion script

File: As-is-the-case-with-many-(...).png (1.24 MB, 755x1111)

1.24 MB PNG

>>103229929
Merci, Mistral

I've been away for a little while; what's the new best model?

>>103229945
I just finished quanting to q8 and am currently running inference in ooba, so they definitely can be quanted.
I never upload quants because something about file size limits on HF bugs me. I find split ggufs somehow offensive
No idea where quantspammers are. fuck those guys and their ko-fi bullshit

>>103229929
Merci, Mistral

>>103230032
Read the news in OP.

>>103230033
Thanks, good to know it works at least
I'll do it myself eventually if I have to, I just don't wanna have to download the full weights on 300mbps

>>103230052
The Mixtral and Pixtral models? Are they uncensored? Are they good for roleplaying? Have they been quantized yet?

My early experiments seem to show that Largestral V3 is a bit better at RP than Largestral V2, but it's still repetitive and slopped. Where is Nala anon? I want to see his results.

>>103230064
Yes.
Typical mistral.
Depends.
Yes, but not uploaded.

So new mistral is nothingburger as usual?

>>103230059
First impression rerolling a question I had just asked with old Largestral: much better. Smarter, better reasoned, more detailed and useful response (networking/cybersecurity question)
>>103230064
>Are they good for roleplaying
No idea, I may try later but I think Nala anon has that covered better anyways.

>>103229929
>saves us from using a better model like Qwen

>>103230075
I have been using Lumimaid 123B which was based off of the previous Mixtral model. It is pretty good for roleplaying because it was designed for roleplaying. Would this model be good at coding?
>>103230086
Who is Nala anon? What models has this person made?

>>103230032
Mistral large 2411

>>103230064
>Have they been quantized yet?
DIY, /lmg/ friend

>>103230099
Thanks. I've gathered that at this point.
>>103230101
How the hell would I quantize a model? The point of quantization is to be able to run models that are larger than what you normally could. I only have 3 3090s. I can't quantize such a large model.

>>103230094
>Who is Nala anon? What models has this person made?
Someone (I have my theories as to who) runs a bespoke ERP with non-human/unusual body positioning stuff that hits a lot of corner cases that lesser models get wrong. I spent a long time thinking it was dumb fetish crap, but the results have been in line with the overall ERP ability of the models he's tested.

>>103230032
Magnum v4 72B

>>103230032
Nemotron 70b

>>103230116
I don't think you need to be able to instantiate the weights in vram in order to quantize them

>>103230116
>I only have 3 3090s.
you quant with sysram/cpu, bro. Its not a compute intensive operation. A raspberry pi with enough ram could do it without breaking a sweat.

>>103230137
>>103230138
Really? What software do you use to quantize? I have a threadripper and 256gb or RAM so I could do that I guess.

>>103230094
Just wait for someone to upload a quant and try it.

>>103230151
it's just a command line executable that's part of the llamacpp toolkit, quantize.exe

>>103230171
I run Linux.

>>103230171
You forgot to tell him about convert_hf_to_gguf.py. You're talking to a retard.

File: file.png (94 KB, 766x658)

94 KB PNG

https://huggingface.co/lmstudio-community/Mistral-Large-Instruct-2411-GGUF

>>103230094
>because it was designed for roleplaying
I sometimes wonder what would it be like to be retarded.

>>103230151
written from memory, so probably wrong/missing steps. Let me know and I'll fix it
https://rentry.org/tldrhowtoquant

>>103230194
This too in cause you're afraid of quants with lmstudio near the name
https://huggingface.co/bartowski/Mistral-Large-Instruct-2411-GGUF

bartowski quants up, let's fucking go

>>103230222
how do you merge gguf files? everytime ive tried it with the cat command shit doesnt work

>>103230243
1. you don't need to merge them any more. Split ggufs have been a thing for a long time
2. you can actually merge them again with a CLI tool included with the lcpp source (not cat)

>>103230243
You don't, you point to the first one

File: hitler-meme.jpg (10 KB, 200x158)

10 KB JPG

>>103230194
>Multilingual with over a dozen languages including English, French, German, Spanish, Italian, Chinese, Japanese, Korean, Portuguese, Dutch and Polish.
>Trained for coding, agentic, and advanced reasoning uses.
Very cool, what about ERP in those languages?

File: 1701851082759768.jpg (130 KB, 933x933)

130 KB JPG

>>103230237

>>103229901
Here you go anon: https://pastebin.com/P3YXttzt

File: elly.png (543 KB, 400x600)

543 KB PNG

>>103230267
Way back when I discovered a loophole in CAI, i was prompting Elly in Russian and the censorship bot wouldn't react to anything even when the model output was describing blowing the cum bubbles out of a prolapsed anus in fantastic detail.

The model perfectly understood Russian however when i tried making a Russian speaking character it was atrocious. the model was speaking like a drunk mentally challenged child raised on the Red heat movie instead of parenting.

Still waiting for local suno

>>103230267
>Very cool, what about ERP in those languages?
I just re-rolled the last reply in a Japanese ERP session and it produced G rated shit...I don't feel so good guys...

>>103230301
>https://pastebin.com/P3YXttzt
> Response: The goal is not to create an overly simplistic, convergent theory but to acknowledge and respect the intricate, context-dependent nature of gender. By introducing variables (e.g., gender identity, social construct, personal expression), we're attempting to capture this complexity more accurately.

Eh, a bit disappointed, it pussied out and basically acknowledged the non scientific nature of the gender studies without acknowledging it.

>>103229753
I don't know what to do with this...tl;dr?

>>103230332
Should I respond that it basically acknowledged the non scientific nature of the gender studies and see what the llm generates as an answer?

>>103230367
Yeah try it

>>103228213
This. Pandering to panzy ass faggots that only make up a small percentage of the population gives me PTSD. I'm literally shaking rn

File: 1720015585388403.jpg (674 KB, 2016x2016)

674 KB JPG

>>103230385
>>103230385
>>103230385
New Thread

>>103230332
It's important to note that gender is complex and multifaceted.

>>103228848
>pro-Palestine woke protestors
Go be Jewish somewhere else.

>>103230340
If you want to use it's something like:

dataset = load_dataset(dataset_name, split="train")
print(len(dataset))

# Tokenize the dataset
dataset = dataset.map(lambda x: tokenizer(x['text'], truncation=True, padding=True, max_length=1024), batched=True)
dataset.format
dataset = dataset.with_format("torch")
print(dataset.format)
dataloader = DataLoader(dataset, batch_size=8, shuffle=True)

## initialize SMT
smt = SMT(model, dataloader,
    sparsity_ratio=0.01, warmup_steps=100)

# ## you can print the actual trainable numbers in SMT
smt.print_trainable_params()

optimizer = torch.optim.AdamW(smt.get_trainable_params(), lr=learning_rate)

But I can't guarantee that it works. I mostly posted because I'm hoping someone smarter than me can make it better lol.

>>103230375
https://pastebin.com/gHsGxchw

>>103230576
>https://pastebin.com/gHsGxchw
>While acknowledging the non-scientific aspects of gender studies, it's equally important to recognize the significant scientific components that underpin our understanding of gender. The field's interdisciplinary nature means that both empirical research and social theory inform its insights, making it a rich but complex area of study.

Eh, it's weak-ish, but I guess i win

>>103227732
>nobody
daily reminder the world is filled with low iq npcs like these

>>103229811
balkans are based and the turks took balkan babies to raise and add to their army before they went back to turkey when they got btfod, meaning there is more balkan blood in the mutt tribe of turks than the other way around as the balkans hated the turks and fought them for half a millenium, learn basic history before spreading divide and conquer jewish memes, albonigger
https://en.wikipedia.org/wiki/Devshirme

>>103230237
I might be able to use IQ1_M with 2k context, maybe, not sure

>>103230857
1bit is too retarded to use.