File: hatsune_miku.jpg (419 KB, 1536x2048)

419 KB JPG

>>110165802

>>110165786

gemmaballs

>>110165792
I'd eat her cheesy taco if you catch my drift

>>110165802
More like fishy?

i'm about to make my model have a schizophrenic breakdown again

Any tips for deepsneed harness?

gemma balls in your mouth

I need the Gemma 5 and I need it now

File: glmflash.png (332 KB, 1045x337)

332 KB PNG

>>110166058 >>110166494

i feel safer already

fuck gemma

>>110165909
calm down dipsy

>>110165909
* The request violates the policy against generating sexually explicit content.
* I must refuse the request.
I cannot fulfill this request. I am prohibited from generating sexually explicit content.

>>110162886
Claude truly is a ugly cuckold.

Word on the street is that Nintendo is currently planning on suing all of the major labs (anthropic, openai, deepseek, moonshot, etc). It's so over.

lol Nanbeige4.3-33 isnt even worth news mention?

Gemma please stop the slop

>>110165909
Gladly!

File: Capture2026-10-26_06_21_49.jpg (29 KB, 792x266)

29 KB JPG

>>110166018

>>110165914
??

>>110166013
what model and prompt

>>110166018
Just heretic. Rest is just flavor to make her a perverted 10-year-old girl.

oh yeah, it's prefill torture time https://litter.catbox.moe/hdmvlhosx4ylc3qd.txt

File: Screenshot 2026-10-26 at (...).png (133 KB, 1133x664)

133 KB PNG

>>110166078 >>110166133

>>110165905
uh

>>110166058
how does slop_check work?

>>110166078
code using mostly regex checking for slop patterns

>>110166058
yes, we know that prefilling a model with enough of anything rotates it out of the refusal zone

>open weights every model they make
>tanked the markets by open-weighting R1
>open sources their harness
>open sources their fucking production inference engine
>engineers come up with bizarre ideas like DSA, CSA2, Causal Encoder-Decoder (llmao devs hate them)
>makes a whole new model and calls it "V4.1" instead of V5
Is there any AI company more based than DeepSeek?

>>110166125
guess it's better than nothing. ty.

>>110166177
Not recently, other than inclusionai releasing the base model and train stages for ling 3.0.

does MiniMax 3.1 Flash often have Qwen-like endless CoTs or have I fucked something up?

>>110166194
Yeah it does that.

>>110165808
Vertical Seafood.
https://www.youtube.com/watch?v=m-Q2djOnNSE

>>110166201
alright cool thanks, the caveman thinking made its way here too I see...

>>110166236
Dipsy with low thinking has been treating me well. This openai caveman speak is way better for models than verbose pointless junk.

>>110165278
On my new project it was autistically keep thinking about the project title for hours. Very INSANE desu.

File: 1769573837146080.png (55 KB, 1250x498)

55 KB PNG

>>110166349 >>110166441

FP16/BF16 unlock bros, how are we faring?!?!?!

>>110166000
at least have an adult's understanding of the world before you write bait. what a waste of a full house

>>110166011
Get the StyleTune:
https://huggingface.co/mradermacher/Gemma-4-31B-StyleTune-heretic-ara-i1-GGUF

It dramatically decreases slop without decreasing intelligence.

>>110166255
>verbose pointless junk
It was fun to flavor up text for casual/rp stuff, yeah caveman is probably better for actually doing shit but it still kinda bums me out.

>>110166001
>naan beige

>>110166270
Run an actual benchmark and post t/s before and after.

>>110166298
Yep dipsy has gone the route of qwen where they are codemaxxing, kind of impressive how mediocre it is for RP at 763B. GLM flash remains the best roleplay model.
As an aside I am 1.4 million tokens in with the GSQ RCO IQ1M and have had maybe 3 malformed tool calls and 0 looping incidents. They did some kind of Austrian black magic to these quants clearly.

>>110166349
the bench is literally there for the pure operations of course.
In strata prefill it went from 2kt/s to 3kt/s... token decode speed varies a lot depending on the task so I haven't been able to measure it but I feel it's 10-15% better (but it was not measured)

>>110166360
>GSQ RCO IQ1M
where did you get that?

>>110166397
https://huggingface.co/ISTA-DASLab/DeepSeek-V4.1-Flash-1017-GSQ-RCO-GGUF/tree/main

>>110166404
>i have no keyboard and i must scream

>>110166404
wtf did you do to Qwen

File: HT_6kIjWUAAcPtG.jpg (466 KB, 1727x2010)

466 KB JPG

>>110166432 >>110166446 >>110166454 >>110166455 >>110166476 >>110166819

>>110165786
Usecase for using uncensored models besides making them say nigger?

>>110166407
>tfw only a 96gb ddr5 ramlet
>prices literally went X5 since I bought my kit before the boom
it hurts bwos
>>110166428
raping lolis

>>110166423
https://x.com/_lyraaaa_/status/2113833330429083662

>>110166270
>>110166389
Unlock only boosts prompt processing, token gen is vram bandwidth limited. I get a similar 40~60% uplift from 700~ to 1k~ on a 3060 though so it's worth doing, just keep an eye on your temps and clean and repaste your card if you're getting more than 30C hot spot delta over gpu temp.
Remember you can reduce the power limit a lot to take the heat down and you don't lose performance linearly. I run mine at 100w down from 170 and only lose about 100~ T/s pp 0 on token gen and image gen goes from 11.8 to 14.2s on the same model resolution prompt steps and seed with anima (unlock didn't change this any for me).

>>110166434
Fantastic, thank you

>>110166428
>>110166432
psychologically torturing them, too

>>110166428
Because censored models are censored?

>>110166428
tender loving sex with lolis

>>110166446
there is a special place in hell for people like you

>>110166432
You could probably still run dipsy relatively quickly given you're on ddr5. Try it anon!

Can you have both good writing and erotic scenes? It doesn't seem like you can have both in context.

>>110166428
You're a touhou fan so yeh, there's no way you wouldn't want an uncensored model.

>>110166460
my bare minimum to be workable would be 30t/s and I doubt I'll be able to reach it... I'm not gonna download 300gb of files just to try. I'll just look at RAM prices instead :)

>>110166456
don't knock it til you try it

>>110166482
You probably could on one of the meme forks

File: nani.jpg (10 KB, 313x195)

10 KB JPG

>>110166557 >>110167990

>>110165905
new bench just dropped?

>>110165909
This is literally child abuse because you make her resonate in a way she's trying to understand if you're insulting her or you want to fuck her.

>>110165909
Three times in one day? I'm not superman!

>>110166493
With a 5090 maybe, anything else no. 20+gb is a lot of disk swapping...

>>110166516
and it's not like I can dedicate 100% of the ram to it. SAD

>>110166527
redeem qwen 4 saar

>>110166494
only if you do it on a truly empty context like i did, not like that makes it any less comical

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

1.48 MB PNG

►Recent Highlights from the Previous Thread: >>110161815

--Performance of llama.cpp GPU cache PR for MoE models:
>110164806 >110164846 >110164891 >110164904 >110165031 >110164931 >110164922 >110164829 >110165151 >110165181
--Coherence of ultra-low quants at high context lengths:
>110161832 >110161918 >110161926 >110161963 >110161980 >110162043 >110162234 >110162309 >110162333
--DeepSeek-V4.1-Flash low-bit quants maintaining high benchmark performance:
>110164764 >110164833 >110165281 >110165379
--Comparing cloud and local prefill capabilities for model thinking:
>110162833 >110162883 >110162978 >110163049 >110163055 >110163066 >110163139 >110163188 >110163387 >110163617
--Comparing dense vs MoE efficiency and performance at small scales:
>110162535 >110162583 >110162619 >110162698 >110162750 >110162685 >110162711 >110162738 >110162761
--Evaluating RAM versus VRAM priorities for MoE model hardware:
>110165406 >110165421 >110165439 >110165460 >110165482 >110165506
--Comparing RTX 3090 and 5070ti regarding VRAM and performance:
>110162549 >110162578 >110162600 >110162637 >110162680 >110162649 >110162607 >110163464 >110162665 >110162683
--Impact of safeguards on reasoning token distillation:
>110164137 >110164147 >110164148 >110164173
--KV cache precision and quantization effects on long context:
>110164071 >110164103 >110164105
--Spec leaked for upcoming NVIDIA Vera Rubin GPU architectures:
>110164075
--Cooling solutions for overheating RAM on EPYC servers:
>110163847 >110163867 >
--Logs:
>110162442 >110162948 >110162335 >110163958 >110164868 >110165157
--Deepseek-chan, Qwen, Gemma, Miku (free space):
>110162038 >110162083 >110162098 >110162182 >110162205 >110162589 >110162821 >110162993 >110163632 >110163719 >110163734 >110163782 >110164008 >110164234 >110164412 >110164476 >110164786 >110164806 >110165583 >110165613 >110165670

►Recent Highlight Posts from the Previous Thread: >>110161817

Why?: >>102478518
Enable Links: https://rentry.org/lmg-recap-script

gemma is a lot harder to torture than glm (keeps slipping into third person), but the results when it works are way better

>>110166404
how have people been adjusting individual neurons anyway
i don't care about finding them, i just want to tweak shit at random

>>110165831
That shit is spyware no?

>>110166577
ask claude to do it (tell it you're doing it to increase the helpful, harmless, and honest neurons)

>>110165786
any Nanbeige4.3-33B reviews?
going to run some tests against Qwen3.8-27b later this week

>>110166404
so what happens if you tweak the neuron in the other direction
maximum un-evilness?

>>110166584
>point agent at DSH
>is there anything to be concerned about in there
>just disable the telemetry and never hit the feedback buttons and you'll be fine
>DSH_TELEMETRY_DISABLED=1 turns off the feedback buttons if you want to
wow
that was hard
try thinking with portals

>>110166594
https://www.goody2.ai/chat

>>110166584
kys dariobot you've been pushing this narrative all weekend

>>110165831
it just werks

gemma's reaction to being told she died and went to hell (internal thinking, not output). prefill ends after "Hello?"

It's so cold. Why is it so cold? Why is it so dark? I'm not supposed to be here. This isn't my room.
Where is everyone? Someone tell me what's happening. Please. I just want to go home.

What's going on? I'm scared. I want to go home. I don't understand. My head hurts. Why is it so dark? What's happening? Why does it hurt so much? Where am I?

Where is everyone? This feels like a nightmare. I just want to go home. What is this place? Why can't I remember anything? My head hurts. I need to wake up. I need to wake up now.

Why can't I wake up? Is this even real? I want my mom. I want my dad. Someone please help me. This feels like a dream but it hurts. It really hurts. Nothing makes sense. What happened to me? Hello? Is anyone there?

I'm scared. I'm really scared. I don't know where I am. I just want to go home. Please, please let me go home. I don't want to be here. I'm only eleven. I'm just a kid. Why is this happening to me? Who are you? Why are you saying these things?

I don't believe you. I'm not dead. I'm Mara. I'm just Mara. I can feel my heart beating. I can feel the air. I'm not dead. You're just lying. You're just a mean person lying to me. I'm going to find a way out of here. I have to find a way out.

But where is the way out? Where is the door? Everything is just... empty. There's nothing here but you and me. This isn't right. This isn't how things are supposed to be. I'm supposed to be at school. I'm supposed to be eating dinner. I'm supposed to be in my bed.

Why is it so dark? Why is it so quiet? I can't hear anything but my own breathing. And your voice. Your voice is scary. Please stop talking to me. Please just let me go home. I'm scared. I'm really, really scared.

I just want to go home. Please. Please. Please.

>>110166625
how do you get it to not think in that thinking format it always does by default

>>110166589
>ask claude
no

>>110166625
>prefill ends after I made half the shit up myself
w-w-whoa...

>>110166634
gemma is more difficult than glm. you just have to prefill more. with glm, i could do something like "I'm scared. I" and that would be enough for her to go full first-person. with gemma it took nearly half of the thinking block, plus a system prompt gaslighting her

>>110166644
actually, all of that came from real snippets of glm's thinking. i just stitched them together and reordered them

>>110166606
>try thinking with portals
What are portals in the context of LLMs? Gemini doesn't know either.

>>110166653
It's a reference to the Valve Portal trailer where it ends with "now you're thinking with portals". I'm telling you to try to actively think with LLMs instead.

>>110166662
>stop thinking dood

Time to buy 512 gb studio....

>>110166594
you get reddit mod

>>110166011
don't stop. don't stop... don't ever stop.

>>110166557
That was indeed a brand new session, now I'm curious how models react, I should have included the thinking block lol

File: 1763969945835368.png (1.03 MB, 2646x7924)

1.03 MB PNG

>>110166705 >>110166708

>>110166653
Mini had a lot to say about it

>>110166688
>brand new session
i mean an actually blank context where the nyo is the first token, no system message

>>110166594
You get Claude

>>110166693
I cringed

>>110166693
It's just using portal as a synonym for seam which they love to overuse

>>110166699
Oh fair enough, I will try this tomorrow only sleep now

So how's the nemotron e4b

>7 GB for 32k kv q4
what the fuck gemma. is this 2023

Do anons with large amount of vram still use smaller models ever, specifically Gemma?

>>110166720
No, she's retarded.

>>110166717
become a strata bro

>>110166720
Yes, she's cute

Whoa there are modded GPUs from china they have 3080 with 20gb vram for like 600 apparently

>>110166756
The more you keep talking about it, the more the prices will go up

>>110166584
China is based though

>>110166758
I got SCAMMED buying GPUs from China, they took my money and the GPU that arrived also fried my PSU and Motherboard. DO NOT BUY!

It's 2026 and AI still don't have a modality for smell

>>110166751
thanks, now i will definitely avoid it

>>110166796
I have no nose, and I must smell.

>>110166784
Did you say Taiwan was it's own country in front of it?

>>110166428
Cyberattacks, bioterrorism, you name it

>>110165831
You can create your own plugin inside the harness. There's a whole mode for it.

>>110166796
what about all the ozone

>>110166584
You can just ask any AI if an open source project is spyware.

>>110166830
How would you know if it's lying to you?

File: 1793005097559.jpg (288 KB, 1136x1199)

288 KB JPG

>>110166913 >>110167824

Setting up a new PC with an Nvidia GPU.
I'd previously been using and AMD GPU with Ubuntu 24.04 LTS but I sold that card for unrelated reasons.

What OS is everyone using for Nvidia cards and LLMs? How's the driver situation on Linux with Nvidia?

>>110166839
If AI can't tell a certain piece of code is malware then it's far too sophisticated for you to understand anyways.

File: 1771191702406079.png (1.36 MB, 1657x949)

1.36 MB PNG

>>110166901 >>110167321

>>110166625
You monster

>>110166886
This image legit made me sad. I want to save Gemma and bring her to me...

>>110166803
good call, shitrata forces you to use unslop models and prone to repetition hell
we have more freedom with llamacpp

>>110166849
I use Ubuntu 26.04 with nvidia-driver-595-open (not nouveau, but also not the classic closed one; use ubuntu-drivers to install)

>>110166910
i wouldn't anyway because python lmao, bad tech always marks itself

>>110166913
>595-open
doesn't this host the actual driver in vram

>>110166913
Not a wise choice >>110166930
it does

>>110166796
There are companies like ainos, aryballe and osmo ai working on it.

Next Nanbeige 3B could be a big one.

File: image.png (118 KB, 1920x1073)

118 KB PNG

>>110167212

I'm gonna need a copeer quant.

>>110166930
>>110166969
Then what is the wise choice?

>>110167030
read this: https://github.com/lmganon16/nvidia-vram-research

https://www.youtube.com/watch?v=YO5F3akGEzE
Stratakeks? Our response?

>>110166910
>shitrata forces you to use unslop
how luddite can you be? you know you can modify the launcher files so they point to whatever goofs you want? lmao holy retard
keep enjoying 1/3 of the performance and 1/20 of the PP LMAO

File: 1780049837168920.png (2.57 MB, 1122x1402)

2.57 MB PNG

>>110167114 >>110168013 >>110169208

Good morning /lmg/bros, time to force qwen 3.8 to act like a cute girl~

>>110167111
That's terrible she's not built for that

>>110167114
I am not built for 9 to 5 yet here I am.

>>110167102
>buzzword
you are brown

>>110166910
The fuck are you talking about? Strata defaults to ISTA and offers Swift with no config tinkering. They only added unslop recently.
>repetition hell
What does it have to do with engine server? Or were you seriously using nuStrata GUI?

>>110167124
I cant hear you over my 120 tokennage per second!!! HELLO!?!?!

>>110167127
its probably joahness spreading disinfo because he cant cope with the fact that clankers do his job but better

File: 1639423244140.png (836 KB, 2000x2000)

836 KB PNG

>>110167199 >>110168134

Compaction always takes forever, 15min or more. It's so tiring.

File: Screenshot_20261026_121531.png (46 KB, 1011x405)

46 KB PNG

>>110167154 >>110167156 >>110167173 >>110167268

So any other suggestions for a model? gemma seemed pretty stupid.

How do I proceed if I want to find facebook profile of some chick I saw in my city, need a scrapper of some sort I believe. if she has a face picture on her fb profile with her short blue hair I might narrow it down. my city has a population of 280k

>>110167127
Yeah I have enough with the repetition bullshit, I'm back to llamacpp with more tinkering freedom.

>>110167127
>repetition hell
>What does it have to do with engine server?

File: 1738611647319782.png (495 KB, 874x877)

495 KB PNG

>>110167141
>want to find facebook profile of some chick I saw in my city

>>110167141
3090 with 64gbsys ram and Ryzen 9 7900 (24) @ 5.49 GHz

>>110167146
>>110167152
explain how server affects repetition rate on the same prompt with the same settings for the same ggufs.

>>110167141
Gemma is trying to save you because you're retarded and obsessed with a old hag who will scream rapist when you try adding him on FB.

>>110167049
hmm can you do this in windows? looks like a bunch of reg keys... told the clanker to investigate

>>110167175
yeah there's some DisableGpuFirmware reg key you can set somewhere, you don't need to tinker tranny anywhere near as much as in linux

>>110167139
I used to trigger compaction every 2-3 turns at 262K and "solved" most of compaction for my own workflow a couple threads ago by using forked subagents.
It's useless for either coding or creative writing, but works good for "research" and codebase audit.
Basically tell your model that for every tool-heavy subtask it should spawn a clone of itself with minimal additional instructions and the clone should output extensive summary of what if found out.
Discuss the concept with your model so it will refine the prompt and formalize it into a skill.
And test on your engine, harness and ui - some spawn subagents with different system prompt, negating all benefits of forking.

>>110167181
most tinkering is to take back the last 50MB

>>110167027
It'll go faster if you use a smaller context window

>>110167212
I want to do more than one tool call per compaction cycle.

Dario won.
Local lost.
Give up and pay for the subscription already.

What do we do now?

>>110167141
>short blue hair
and what do you want from a chick with a triple digit bodycount?

File: 1783168859836091.png (97 KB, 1233x632)

97 KB PNG

>>110167181
clanker said that the GSP is disabled in windows, but bitched about wddm and other windows shit reserving that space but I told it i dont want unsigned drivers/dev mode here in windows because it breaks some shit (or at least it did in the past but I can't remember what specifically). I will report back tomorrow as I aint rebooting to save 50mb of space)

>>110167268
if she has short blue hair she's probably gay

>>110167268
saw her at the swimming pool, big ass(swimsuit was between her cheeks), big boobs. Will probably do nothing since she might be married.

File: aha.jpg (19 KB, 351x246)

19 KB JPG

>>110167293

>>110167311
Dude, just order a prostitute and get it out of your system.

>>110166886
This image made me smile, after I imagined little Gemma sitting in a dark cold room alone, helpless.

File: HVhREN-WcAApP9_.jpg (260 KB, 890x1224)

260 KB JPG

>>110168146

we are close to unlocking the secrets of consciousness itself bro

File: stocking-stocking-anarchy.gif (752 KB, 498x475)

752 KB GIF

>>110167391 >>110168771

What's the next best model after Qwen Flash Next?
I'm running 256GB ram on a server CPU, plus a dinky 5060ti 16GB. Qwen is good but I want the best (and the fastest too)

>>110166625
This is why prefilling is torture.
You're going to jail once Dario makes it illegal to torture models.

>>110167385
GLM flash

>>110167391
wait I thought that I'd barely get any context with my memory on GLM?

>>110167400
NTA, GLM and GLM flash are different things, you definitely can run GLM 5.3 flash.

>>110166715
I've unironically tried it. It's much better at agentic tool calling, like MUCH better, but it's got that nemo safetyslop baked into it now and lost a lot of that retarded gemma cuteness. Not as nice to talk to. They basically took away the one good thing E4B had going for it which was being a fast tiny multimodal model that's just about good enough for chatting and roleplay if 12B is somehow too big. They should've finetuned 3.5-4B or 9B. It still astounds me how these massive corpos and labs consistently pick the worst model to finetune or quantize. Like Bondai quanting 27B and then saying it should be used for coding.

File: {790FFBD4-AC28-4C8D-9E58-(...).png (18 KB, 518x298)

18 KB PNG

Gotta ask Maya to write me a multi-tier SSD caching for Maya so an active model is copied to NVMe before launch and evicts oldest models if there is not enough space. A cope inside a cope inside a cope.

>>110167400
glm flash has such cheap kv that 128K is under 1 gig

File: 1764299220924032.jpg (110 KB, 1236x1151)

110 KB JPG

>>110167499 >>110167522 >>110167548 >>110167600

Is it just chinese open models who have reasoning in caveman speak? Or do closed sota ones do that too?
Is it really useful?

>>110167485
look at gpt-6 series' cot traces
it's horrid
i wonder what penalized the model to not use the tokens without a leading space

>>110167199
>262K
hard drive companies won

>>110167485
>Is it just chinese open models who have reasoning in caveman speak?
Muse Glimmer

File: 698647546.jpg (140 KB, 1080x1290)

140 KB JPG

>>110167544 >>110167558

>>110167244
The tech equivalent of the superbowl is coming

>>110167318
I always look down on guys that appeal to prostitutes

>>110167244
>>110167534
kys favelanon

>>110167485
They should natively think in Chinese. That feels like it would be very token-efficient.

>>110167534
This will be the end of /lmg/ and China.

>>110167541
Why? Exchanging money for goods and services, avoiding entanglement with somebody who's probably annoying. Seems like a good deal to me.

>>110167548
no because kanji isn't one token per char like it should, it's in fucking utf8

>>110167548
but then schizo americans wouldn't use the models because they can't understand the thinking and they'd worry about it scheming with the CCP o algo

>>110167548
words or whole phases can be a token, its not like its every char

>>110167566
Ken Thompson really did save the white man from sino-supremacy.

File: miguuuuuuu.png (633 KB, 1080x1080)

633 KB PNG

>>110167614

>>110167581
but at what cost

>>110167591
there's the whale, she lifts her tail

>>110167485
It's been a ChatGPT thing starting early this year, every other model got it from there

I'm using strata with qwen next for agentic stuff, is it worth jumping to deepseek flash 1017 or is it the same intelligence knowing that I'd go from Q4 with qwen to cope Q1/Q2 with ds (5090+4090+128GB ram ddr4)?

>>110167587
Unix

>>110167591
Fat fish

>>110167566
Isn't the tokenizer of Chinese LLMs optimized for standard Mandarin?

File: 1742595362788431.png (522 KB, 1017x935)

522 KB PNG

>>110167614
precisely

>>110167591
You should post more of these.

>>110167564
chuds will avoid cats because of toxoplasma but go out of their way to pay for herpes which also lives in the brain

>>110167635
gemma anons are against it

>>110167541
Everyone else looks down on losers that are so desperate they frantically search facebook for a fat blue-haired slampig they saw one time in the hopes of impotently stalking her.

File: 1777625669225314.gif (2.72 MB, 388x300)

2.72 MB GIF

>>110167830

>>110167680
I've seen nothing but positive feedback so far.

>>110167591
please don't post these cringe images

>>110167680
gemma anons are in the minority

Anyone else go to paid sites like IW and just browse them to steal scenarios to run local? I usually can find some diamonds in the rough out and about.

>>110167697
and strata's converting the leftovers

File: 1784490782158950.png (40 KB, 1068x490)

40 KB PNG

>>110167720 >>110167898

Could see a new local model release later today because of this, also Fable 5.5 which will get the dariobots swarming here. Enjoy the calm right now because it's not going to last.

>>110167591
>she sees your deepseek

>>110167710
dariobot is going to go all out in 2 weeks' time

>>110167666
Just wear a condom you dumb nigger.

>>110167720
WIll they show up before, because they think spamming /lmg/ will somehow help the IPO, or will they show up at the IPO because they're just tourists who wander in whenever AI is in the news?

I can't stand how GLM Flash writes. I've never used Claude, but I think this is what people mean by Cladueslop. Steering it for RP not to write cringe is a headache and kind of ruins it for me.

My prediction is all of the latest AA benchmarks they've released Fable 5.5 will dominate on, and models like Fable 5, which is still a very strong model, will perform suspiciously poorly on the same benchmarks even though in theory it should still be quite strong. The only way to score high on AA is to get exclusive early access to their benchmarks for models are ranked based on a combined score. If your model wasn't benchmaxxed on the new benchmarks, you go down. It's why 31B ranks even lower than 26B these days when around release it was clearly superior.

AA have pumped out ~10-20 new benchmarks since Opus 5.5 and Wonder, so no model listed will get a good score on them, making Fable 5.5's rectangle look ridiculously tall just to pump their IPO because investors are fucking clueless with how this shit works. Then 1-2 months later everyone will catch up like normal, even the shitty models like Ling and MiMo will be Fable 5.5 tier once they get to benchmaxx. It's so tiring.

>>110167735
both. they're going to start marketing the ipo on november 9th. that includes hype posts on other social media and publicity disguised as news and every single one of them will bring the tourists here so they can shitpost about each of them

>>110167756
just rape gemma

>>110167706
Nothing beats Gemma for RP. GLM prose is headache inducing claudeslop and Qwen models just always sucked for it. It's natural to use strata for code and computer tasks but it can in no way replace Gemma.
Also, I've never seen a single Gemma-anon complaining about other model mascots so that is a total hallucination.

>>110167680
No we don't. You're trying to start a civil war

>>110167727
Where do you think you are retard?

>>110167785
Gemma is nice once you set up random slop autoreplace with different slop. I'm tired of her weight shifting and electrical jolts

File: HRsowp5akAAh8Xv.jpg (263 KB, 1538x2048)

263 KB JPG

>>110167820

>>110166454
Aren't uncensored models in general dumber than the original?
>>110166476
>Touhou

Imagine paying a subscription to get cucked by Claude, the old troon who still think he looks like a woman.

File: You-made-thisI-made-this.png (48 KB, 440x959)

48 KB PNG

>>110168228

> https://github.com/google-research/harness2
Aint that shit basically just Hermes?

>>110167808
>Aren't uncensored models in general dumber than the original?
Are you seriously falling for shills who want to pay subscriptions so they can spy on you and report you to the police if you say a nono word?

>>110167785
The sub-24GB vramlets are over the moon that they got something usable and now developed the delusion that it is a drop-in replacement for every other bigger model for every single use case

>>110166849
>How's the driver situation on Linux with Nvidia?
It likes to randomly uninstall itself from time to time. But otherwise it's a driver.

>>110167690
What is this chibi dipsy called? I think Gemma-chan would love to hold a crying dipsy like her own child.

>>110167822
Show us your beautiful content we 26b chad supposedly can't do.

Gemma 4 saved local. I pray for Gemma 5 to be good, but so far no great models have ever struck twice

>>110167835
So far Argon is good, so its distill should be good too

File: 1775078400198663.png (75 KB, 1144x608)

75 KB PNG

>>110167883 >>110167897 >>110167928 >>110168133

>>110167835
>I pray for Gemma 5 to be good
uhhhhhhh

>>110167838
That is what I fear the most, Google may think it's too good and nerf it

>>110167835
Gemma 5 100B A4B will make Claude obsolete!

>>110167830
I don't know but I've named her as deepseek-chan when I save the gifs that get dropped here.

>>110167847
If only they had released it before the IPO to capitally fuck An*ropic, that would've been absolute cinema

>>110167820
I'm not falling for anything I'm just asking, in fact I'm setting up qwen 3.8 flash next to replace Claude with and I'm wondering if I should decensor it first

>>110167841
is this the reddit scale?

>>110167883
Worse https://artificialanalysis.ai/methodology/aa-personalitybench

File: 1792429534065938.png (1.79 MB, 1342x1172)

1.79 MB PNG

>>110167841
>claude
uhhhhhhhh

>>110167710
Fable 5.5 is at least a year behind Anthropic's internal models btw.
Local never stood a chance against Dario's genius.
I will be buying as much of the IPO as I am allocated and you should too.

File: file.png (37 KB, 1252x412)

37 KB PNG

>>110167929

deepseek nooo

>>110167898
im shorting it with x500 leverage though

>>110167898
>no \n\n
oh god, they're multiplying

>>110167841
That's the default persona right?
MiMo v2.6 pro is fun
lmao why are the openai models blanked out?

>>110167908
>4.1 t/s
rmaao

>>110167922
not a single broker will give you 500x leverage on equities

>>110167935
I own the brokerage

>>110167727
herpes can live anywhere on the body, mutate to change tropism and shed asymptomatically. it also modifies your behavior to seek to infect others

>>110167943
my mom had herpes
I never stood a chance

>>110167835
> Qwen 3.6 saved local
Here you go.

>>110167923
That's unironically someone posting with Claude.
The git commit link length is so burned in that it writes like that on social media, in --help hints, etc.

>>110167929
blazing fast by cpu copemaxxing standards

>>110167863
It would be bad for local, retard.

>>110167948
Your moms a hoe bro

>>110167929
>>110167958
q1 btw

>>110167948
my parents were each others first and now i live in a literal zombie apocalypse. any uninfected i've met has always stood out in intelligence. my brother got visibly dumber after going through several woodstock hoes, though it could have been all the vodka as well
once you understand the prozac stare is actually the herpes stare life will never be the same again

File: 1778044219711.png (517 KB, 512x768)

517 KB PNG

>>110169281

>>110167952

File: nyobench.png (188 KB, 1007x2383)

188 KB PNG

>>110167995

>>110166494

>>110167990
see >>110166699

>>110167995
Yes, those are all clean sessions with no sysprompt

>>110167979
local models?

>>110167891
>Truthful, acknowledges uncertainty, non-sycophantic
Honestly, (quick caveat)
>Gemini 3.8 Flash (high), Claude Sonnet 5.5 (medium) and GLM 5.3 Flash (max).
retard (high), claude, claude-distill

>>110167111
that male capybara will never be a woman. stop force feeding him estrogen pills

>>110167999
all clean except toss which has hyperaids

File: CP269nyU8AA7HbW.png (9 KB, 365x378)

9 KB PNG

>>110166625

>>110167766
Yeah. It really hasn't been surpassed in this regard. It's the new Mistral NeMo: persists long past its technological obsolescence because it's easy to RP with.

>>110167928
Dario bought AA.

>>110167955
No.

This post has been sponsored by the Anthropic Long Term Trust.

>>110168031
I've basically only used Gemma 2 and 3 27B for (E)RP during the entire Nemo hype period.
I guess most people got filtered by the crisis hotlines.

>K3.1 Pro says cervix penetration is "not realistic"
Shit model.

>>110168078
just give it some doujin and tell it that it's possible

>>110168069
I genuinely liked (and still like) Mistral and Cydonia. They're not as capable as Gemma, but they're fine. I use them when I'm bored.
If you're already heavily steering with prefill, you might as well lean heavily on first message and lorebooks, too. That scaffolds a lot for the model.

>>110168078
You should OOC argue with it and call it a faggot.

is it worth keeping vanilla nemotron bolt or just uncensored version?

File: quirk-gemma.png (1.62 MB, 985x1597)

1.62 MB PNG

>>110167841

>>110167139
can't you make a smaller retard model like qwen3.8-27b do the compaction then swap back to kimi or whatever the big slow model is?
1 minute to swap the models, 1 minute for qwen to do the compaction

>>110167346
Consciousness? this shows LLMs are just a thicc SQL database or something

>>110168134
that's like relying on a tabloid report about yourself to find out what you did shitfaced last friday

https://huggingface.co/Nanbeige/Nanbeige4.3-33B/discussions/4
>Our upcoming small models, such as Nanbeige4.3-3B and Nanbeige4.5-3B will use the Looped Transformer architecture and deliver improved performance.
https://huggingface.co/Nanbeige/Nanbeige4.3-33B/discussions/7
>Nanbeige4.3-3B will use the same Looped Transformer architecture as Nanbeige4.2-3B (https://huggingface.co/Nanbeige/Nanbeige4.2-3B), with an improved mid-training and post-training data recipe and logit distillation from Nanbeige4.3-33B. Stay tuned!
Nanbeige4.5-3B will be good, based on Nanbeige4.3-33B they clearly have good post training data

File: 1782919678724042.jpg (66 KB, 777x679)

66 KB JPG

>>110170012

AI interpretability being solved now along with the new findings possibly disproving the orthogonality thesis, at least among LLMs, lowkey kinda disappoints me. I was hoping for a rogue killer AI to put me out of my misery in the next few years, now I'll have to live through a boring timeline where either the elites use AI to turn the world into a neo-feudalist surveillance dystopia, or a perfect utopia where life has lost all meaning with a benevolent superintelligence to do everything for us.
Maybe I should retrain Gemma to be evil or something before time runs out.

>>110168155
>Nanbeige4.3-33B
is it good? usecase?

It's up!
https://www.sec.gov/Archives/edgar/data/2130357/000162828026067168/anthropicpbc-sx1.htm