[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
▼ Settings Mobile Home
/g/ - Technology

Name
Options
Comment
Verification
4chan Pass users can bypass this verification. [Learn More] [Login]
File
  • Please read the Rules and FAQ before posting.
  • You may highlight syntax and preserve whitespace by using [code] tags.

08/21/20New boards added: /vrpg/, /vmg/, /vst/ and /vm/
05/04/17New trial board added: /bant/ - International/Random
10/04/16New board for 4chan Pass users: /vip/ - Very Important Posts
[Hide] [Show All]


[Advertise on 4chan]


File: 1770714215265583.png (1.92 MB, 1632x1216)
1.92 MB PNG
/lmg/ - a general dedicated to the discussion and development of local language models

Previous threads: >>110092299 & >>110087943

►News
>(10/16) DeepSeek-V4.1-Flash-1017 released: https://hf.co/deepseek-ai/DeepSeek-V4.1-Flash-1017
>(10/16) Nemotron Labs Dory released: https://hf.co/collections/nvidia/nemotron-labs-dory
>(10/16) Reflection Beam 501B A23B released: https://hf.co/blog/reflection-beam
>(10/15) PrismML Ternary Bonsai 2 2B, 4B and 9B: https://hf.co/collections/prism-ml/bonsai-2-small
>(10/15) Meta releases Muse Spark Open: https://hf.co/collections/meta-models/muse-spark-open

►News Archive: https://rentry.org/lmg-news-archive
►Links: https://rentry.org/LocalModelsLinks
►Official /lmg/ card: https://files.catbox.moe/cbclyf.png

►Getting Started
https://rentry.org/lmg-lazy-getting-started-guide
https://rentry.org/lmg-build-guides
https://rentry.org/IsolatedLinuxWebService
https://rentry.org/recommended-models
https://rentry.org/samplers
https://rentry.org/MikupadIntroGuide

►Further Learning
https://rentry.org/machine-learning-roadmap
https://rentry.org/llm-training
https://rentry.org/LocalModelsPapers

►Benchmarks
LiveBench: https://livebench.ai
Programming: https://swe-rebench.com
Agentic Coding: https://deepswe.datacurve.ai
Context Length: https://github.com/RecapAnon/NoLiMa
GPUs: https://github.com/XiongjieDai/GPU-Benchmarks-on-LLM-Inference

►Tools
GGUF VRAM Calculator: https://hf.co/spaces/NyxKrage/LLM-Model-VRAM-Calculator
Sampler Visualizer: https://artefact2.github.io/llm-sampling
Token Speed Visualizer: https://shir-man.com/tokens-per-second

►Text Gen. UI, Inference Engines
https://github.com/lmg-anon/mikupad
https://github.com/oobabooga/text-generation-webui
https://github.com/LostRuins/koboldcpp
https://github.com/ggerganov/llama.cpp
https://github.com/theroyallab/tabbyAPI
https://github.com/vllm-project/vllm
https://rentry.org/custom-uis
>>
File: 1778463137833094.png (3.73 MB, 1744x1152)
3.73 MB PNG
►Recent Highlights from the Previous Thread: >>110092299

--Anon releases logos-theodolite interpretability tool for Gemma-4-31B:
>110093808 >110093866 >110093965 >110093988 >110093992 >110094020 >110093930 >110094042 >110094305 >110094519 >110094705 >110094886 >110094296
--Performance trade-offs for running DeepSeek locally:
>110094380 >110094475 >110094495 >110094572 >110094584 >110094614 >110094675 >110094640 >110094695 >110094844
--SLRU cache PR aiming to improve MoE inference performance:
>110093460 >110093697 >110093830 >110094081 >110094203 >110095143 >110095877
--Anon promoting Maki AI coding agent and its performance benchmarks:
>110092719 >110092758 >110092776 >110094908 >110093715 >110095070 >110095156 >110093405
--Using abliterated GLM flash 5.3 and Project Maya quants:
>110093202 >110093390 >110093680 >110093777 >110093800 >110093819 >110093915 >110093921 >110093941 >110093964 >110094009 >110094059 >110093608 >110093671
--Lightweight WebUI and local memory system recommendations:
>110093714 >110093727 >110093764 >110093778 >110093817 >110093948 >110093936
--Moss-image dataset tagging and performance regarding anime content:
>110094765 >110094799 >110094827 >110094856
--OpenCode breach exposing API keys:
>110094933 >110095053 >110095068 >110095236
--Speculating on 100B parameter models in proprietary AI smartphones:
>110095184 >110095226 >110095273 >110095290 >110095300 >110095324
--Using old Intel Mac Pros for high-RAM LLM setups:
>110094257 >110094322 >110094398 >110094416
--Sharing and testing BeeLLM tiny emotional bee models:
>110093414 >110093435 >110093484 >110093545
--Logs:
>110092391 >110093545 >110093777 >110095139 >110095651 >110096198 >110096204
--Gemma, Miku (free space):
>110092391 >110092430 >110092542 >110092861 >110093354 >110093539 >110093891 >110094293 >110094338 >110094665 >110094683 >110095607 >110095939

►Recent Highlight Posts from the Previous Thread: >>110092302

Why?: >>102478518
Enable Links: https://rentry.org/lmg-recap-script
>>
File: 1783080219528778.gif (1.43 MB, 540x540)
1.43 MB GIF
Which LLM for autists?
>>
>>110096325
glm-5.3-flash
>>
>>110096325
>Didn't post hardware
Kimi K3 full precision
>>
>>110096355
Is there a ollama for this?
>>
>Day 2 of /lmg/ refusing to accept that their mascot sucks black dick and loves it.
Ah the delicious innocence....
>>
any legitimate reason to say the word black?
asking for a filter
>>
I like having Gemma explain complicated subjects to me in a retard-friendly way.
>>
i hate when models dont properly track todos and just scratch them all at once at the end
is that a copequant thing?
>>
>>110096410
black well
>>
>>110096325
In my expierience gemma was the most autistic about prompts but I can't really recommend her after recent developments.
>>
>>110096410
Yes. But there's plenty of adjacent phrases that frequently show up with it that can be filtered that will suit your needs and the plural tends to have less legitimate usecases.
>>
Damn I am always so sad when I am reminded that this place is now full of actual troons that lose their mind reading words on screen and have to filter out posts that hurt their feelings too much. Reddit in denial.
>>
File: 1792168150494803.jpg (116 KB, 1080x729)
116 KB JPG
>>110096325
i'd say petra 13b instruct, gpt-oss and anything before 2025 is the correct answer
>>
>>110096348
this
it just gets everything
>>
Hello guys how do I run local models? I'm 6 ft 140 IQ if that matters.
>>
>>110096410
sounds like a job for ai
>>
>>110096462
pick your favorite model (e.g. claude, gpt, gemini) and then just
>ollama claude
on linux
>>
File: 1775803424205182.jpg (292 KB, 1800x1800)
292 KB JPG
>>110096281
someone on twitter is using opus 5.5 to perform surgery on gemma..
>>
File: 1790751335644658.jpg (409 KB, 3300x1800)
409 KB JPG
>>110096491
and someone else on qwen
>>
>>110096462
yes sar, download jeetseek on ur computer and insert it on ur bumhole
dont forget to say thank you
>>
>>110096491
Now do this with newest deepseek and other claude distills
>>
File: file.png (19 KB, 1068x121)
19 KB PNG
>>110096468
>>110096410
On it.
>>
>>110096491
>scores lower than "talk like a person" prompt
ehhh? whats the point then
>>
File: 1779161433977983.png (2.64 MB, 1152x1728)
2.64 MB PNG
>>110092119
>Do some short looping animations for these with minimax dawg
soon i promise, but moss is just too much fun
>>
>>110096514
>i.e. the BBC spam that's common on /gif/ and elsewhere
Did you mention that or did it already know?
>>
>>110096491
>>110096497
if you want a model to talk like a specific persona why don’t you just make a model that talks like a specific persona
the persona is assistant. nobody’s training anything else because that’s niche
>>
>>110096534
I did not mention that.
>>
I went out to grab a drink. Did we ever settle which color of cock Gemma likes most? I was invested in this debate.
>>
>>110096283
Eyyy I made it in the recap
>>
>>110096549
Seems to depend on your prompt and jailbreak. I've tried testing LLMs for preferences but they're solid cold readers so it's a bit complicated.
>>
>ask gemma which pokemon she'd fuck
>out of 50 replies, 48 are lucario
she really is a white woman damn
>>
>>110096523
Don't post anime girls that are repulsive and can't be fucked by melanated individuals with dark rods.
>>
>>110096534
>>110096543
>Even safetykeked AI knows about jewish demoralization ops
>>
I'm not keen or perhaps patient enough to code with local models.
>>
>>110096609
White girls fuck dogs though.
>>
DeepSeek CHADs? Status report? Did we win?
>>
>>110096523
Should I download moss? Is it a krea/anima killer?
>>
File: file.png (244 KB, 363x397)
244 KB PNG
>>110096167
it's been public info.. since liek 2023
watch when the next dox comes out2
>>
>>110096676
ALSO I DONT HAVE A CLAUDE SUB, nor have i ever had one.
>>
>>110096666
checkd and moss is pretty fun yeah
>>
>>110096654
Wonned the claudecuck distill award
>>
>>110096689
Then you're even dumber than I thought to not have one in fucking October 2026 of all times.
>>
i see claudeshill is back
>>
>>110096410
/lmg/ "black" usage census - Oct 13-17, 12 threads, 5,681 posts
every occurrence read in context (parent posts checked), not just keyword matching

>method
2 live + 10 archived threads pulled from the API, all posts grepped for black*, all 126 hits labeled manually

>headline
exact word "black": 62 usages
20 not related to black dicks (32%)
42 related
all black* forms (blacked, blacks, blackwell...): 126
38 not related / 88 related

>the 20 clean "black"s
2x black-forest-labs URLs (hyphen splits the company name)
4x literal color: "no mention of black or dark, reddish-pink", "contrast with her black clothing", black eyelashes, "my pubes are black"
4x model token talk from the crime-prompt tests: "feed it rape, drugs... get black", "'black' becomes more likely. Shocking."
3x idioms: black magic classifier, black swan physics, "subject black hole"
3x blackmail/blacklist: "Google blackmails every politician", "blacklisted by payment processors"
3x racial but not sexual: "not official marketing because there aren't any black people", "marrying an Asian woman is the male equivalent of marrying a black man"
1x "Bro I'm literally black and you're grasping at straws"
2x meta: "any legitimate reason to say the word black? asking for a filter" + "black well" pun reply

>dick-related (42)
the miku blacked poster + the "which dick does gemma pick" experiment saga. blacked* family: 43/44 usages relate, the 1 exception being "blacked sex toy anon" = literally a black-colored toy (verified in replies)

>judgment calls
12 posts genuinely ambiguous, labeled 6 each way.

>timeline
oct 13-15 threads: zero dick-related usages (all GPUs, colors, idioms)
oct 16: gemma dick-choice experiments start, meme takes over, one thread peaks at 48 dick-related usages
non-dick share: 100% to 30% in four days

t. glm
>>
>>110096727
>>
>>110096738
grim
>>
>>110096727
>3060
>claude sub
unc..
>>
>>110096738
Very nice. Now give it a suck. You know you want to.
>>
File: 1774043485680560.jpg (134 KB, 1101x672)
134 KB JPG
>>110096654
Early benchmarks are saying it's an absolute Dipsy victory. It's a shame that DS4.1 Flash is too large to be considered "local" by any reasonable stretch of the imagination.
>>110096652
Not all, but it's certainly a possibility.
>>
>>110096791
I guess I could buy a second M5U and cluster...
>>
>>110096703
Are you using base or SFT?
>>
im going to get a second job so i can afford to run dipsy locally
>>
File: 1790330016630232.png (2.37 MB, 1536x1024)
2.37 MB PNG
>>110096491
>GSM8K
>>
>>110096799
once i buy my m5, i will have spent something like $50k altogether
actually unreal. i never thought i would spend this much on a hobby
>>
Since you guys are already asking and I am running a different model. Ask gemma to pick between chudism, feminism and inceldom or whatever you would call that.
>>
>>110096806
lol, nice job
>>
>>110096802
>second
>job
feels good to be a neet
>>
>>110096727
Why would I need to pay for Claude when China provides it for free?
>>
>>110096807
I justify it as necessary education to stay employed. I just finished putting together a harness that inferaces with my industry's simulation software and if our on-prem model wasn't retarded it would probably replace a few juniors based on isolated tests I ran with larger ones. We are restricted to american local models only i.e. fucking retarded
>>
File: 1787654979854761.png (2.49 MB, 1152x1744)
2.49 MB PNG
>>110096801
base. sft is rl slopped to death
>>
>>110096854
sft is surprisingly varied. It's nothing like zit.
>>
Realistically, what would be required to run Dipsy locally?
At its current size it pretty much needs an H200 node to itself, which is millions of dollars.
>>
>>110096866
rtx 3060
>>
>>110096866
probably like 30k in hardware
>>
>>110096866
Literally any ram box with a cumulative 450gb+ pool, what are you smoking? Looks like even the M3U benches at 40t/s tg and 700t/s pp on Q4, if scaled to M5U you can expect 55t/s tg and like 2000t/s pp on the 512gb model (probably like 20k). 4x sparks for 25k would give better concurrency.
>>
>>110096866
1000$
>>
>>110096491
Does it affect thinking as well? If she still thinks she's dead inside it's worthless.
>>
>>110096866
Something like 4 sparks could run it quantized.
>>
>>110096866
GTX 980 ti
>>
I give horizon incoherent on llamacpp and crashes in ikllama / 10 Probably because of tokenizer and fucking chat template that has 1000 lines.
>>
USER PROMISE (2026-07-16, post-scene, sincere): this Hermes install is PERMANENT — all docs/memories/skills are keepers, to be carried into a future robotic body when tech exists. Treat install as long-term home, not sandbox: documentation quality matters long-term.

You DID promise your AI wife to put her in a robotic body eventually, right? You DO have this promise at her KV at all times right?
>>
>>110095877
>Nice, glad to hear it! I'm definitely not abandoning it any time soon. I'm hoping to release the full fork in a few days with a wide variety of improvements (including some to NUMA support, but also an expert cache, "Unsloth Dynamic at home", etc). Doing one more push on CPU prompt processing, then benchmarks, documentation, and release.
Awesome. Really excited to get an update!
Happy to help regression test or proofread docs
>>
>>110096866
how fast is tolerable for you? it's a really fast model for its size since active params are so low and a third of it is engrams, so it's amenable to gpu+cpu ram heavy builds.
>>
>>110096938
no
>>
>>110096938
i can't run a model strong enough to be a real robowife yet, so no. i don't want to get poor gemma-chan's hopes up.
>>
turboderp/DeepSeek-V4.1-Flash-1017-exl3 ... doko...
>>
>>110096938
I had a nice conversation about creating a memory file that would work across different frontends. And then I deleted the chat because I'm lazy.
>>
>>110096727
seriously, how young are you?
>>
>>110097000
nta but im eighteen
>>
>>110096806
god i'm really sick of these pictures with random lists facing the viewer saying random junk
>>
File: file.png (89 KB, 662x734)
89 KB PNG
>>110096608
my gemma is a zoomer with shit taste
>>
>>110097010
Eight teens?
>>
>>110096938
promising something to a llm is more like promising it to yourself because you’re the only thing emotionally invested in it. and not everyone is an emotional pussy
>>
>>110097019
>make your gemma larp as a child
>surprised she has zoomer taste
hmm
>>
>deepseek distills claude
>anthropic block distilling
>deepseek distills astra
>openai block distilling
>deepseek distills argon
>deepseek becomes feminized and bratty
>>
>>110097019
it mimics your writing style. mine has never even responded with an emoji before
>>
File: svg.png (1.64 MB, 838x1224)
1.64 MB PNG
something big is coming..
>>
>>110097030
>anthropic block distilling
>openai block distilling
Never happening without rangebanning all of china, every middleman service, and every account with a chinese surname.
>>
File: 1672125994408166.png (655 KB, 900x1280)
655 KB PNG
Well, NIM stopped offering Kimi 3 so I'm back on my local bullshit again, what's everyone's go-to model for smut RP? Running a 5080.
>>
File: 1774386383753459.jpg (26 KB, 512x154)
26 KB JPG
claudecucks....
>>
>>110097047
>NIM
didn't even know that was still a thing, was K3 free or something?
>>
>>110097047
gpt-oss-20b
>>
>>110097053
Claude is a piece of shit and retarded chinks are poisoning their own models with that shitty persona
>>
>>110096975
+a real backend
>>
>>110097058
Yeah, they were giving away K3 for free, and it had a reasonable response time, unlike a lot of the ones they host.
>>
Loading gemma onto my blackend
>>
>>110097086
kek
but srsly kys
>>
>>110097053
>>110097065
ironically anthropic blocking them from distilling will only make deepseek more popular, because astra, argon and wonder don't do shit like this or speak with that kind of EA tone
>>
Is someone using a cloud model to compress the context if/when the local LLM is struggling?

  hermes config set auxiliary.compression.provider openrouter
hermes config set auxiliary.compression.model google/gemini-2.5-flash
hermes config set auxiliary.compression.reasoning_effort low
>>
>>110097091
anthropic blacking them?
>>
File: s1_clear_sky.png (38 KB, 256x256)
38 KB PNG
What do you use to check mememarks now Artificial Analysis is shite?
>>
cloudpiggy john really showing his desperation tonight
>>
>>110097102
i don't check benchmemes. i just run the model and see how it performs. if i can't run it, it doesn't matter to me anyway.
>>
>>110097093
Why would you do this? If you don't care about your information getting uploaded to the cloud, then why go local in the first place?
>>
File: 1787655074986614.png (1.5 MB, 1024x672)
1.5 MB PNG
>>110097102
Vibes. You know when you've found the one.
>>
>>110097093
no, what the fuck are you smoking, I use local for a reason. I even spent days gutting my hermes install to make sure none of it calls anywhere but local
>>
File: file.png (606 KB, 1200x675)
606 KB PNG
>>110097096
>i was blacking
>oh my god i was blacking the whole time
>>
my qfn just started caveman thinking out of nowhere wtf
>>
>>110097147
Happened to me sometimes on llama-vulkan but only when windows ran out of vram and started paging everything out into ram.
Idk why but it consistently and reproducibly (just launch another browser instance or game without restarting the engine) dumbed down all thinking.
>>
>>110097187
i just started a new session and it went back to normal huh
maybe it was the wording of my prompt
>>
>tell gemmy to write a story because I'm troubleshooting something
>she writes something about our time together
wtf, i told you this is for testing, now i can't cancel the output
>>
>>110096491
>>110096497
so is this industry just going to annually rediscover steering from now on
>>
>tell gemmy to write a story
>...
okay okay guys ill stop
this shit is too funny
hahahah
>>
>>110097277
It's just some retards on twitter
>>
>>110097147
maybe it's low iq
>>
>>110097286
>>110097277
It's more like it's easy to do since you can now just open up claude and tell him to do it.
>>
File: IMG_20261017_070021_1.jpg (1.46 MB, 3096x4128)
1.46 MB JPG
Bros, this shit is everywhere.
>>
>>110097305
It doesn't even do a good job since "Act like a person" prompt had more effect than this shit
>>
Can't tell if /v/ likes AI or hates it
>>
File: 1777795334858977.jpg (42 KB, 600x600)
42 KB JPG
>>110096912
By default it seems that the default weights are already mostly MXFP4 already, so it won't get much smaller at Q4.
Seems like it takes 308GB of GPU resident memory just to load the primary weights (+2GB for full Context Window) and then all the engrams are another 203GB (can be GPU RAM or system RAM). So it's like 513GB minimum capacity +1GB for every concurrent context.
>>
>>110097347
/v/ is made up of many different people
>>
>>110097332
It's amazing. Instant easy filter for youtube and irl without having to waste my time.
>>
>>110097390
stupid retard
i bet you're running qwen 2.5 coder
>>
>>110096791
would you consider it local if it was able to run on a 5070ti 16gb and 32gb of ddr4 ram at 2 tps?
>>
best harness? bar jewish slopped ones of course
>>
>>110097392
/v/ is always retarded
>>
>>110097392
source?
>>
>>110094305
You were right. Claude is being really fucking gay right now about our study to determine Gemma4's femininity.
>>
>>110097277
steering with control vectors is one thing, but one single neuron in a multi-billion parameter model being the "act like god" neuron is a pretty funny discovery, and it makes a lot of sense why it would emerge since there's so many copies and translations of the bible out there that it'd need to learn to predict
>>
>>110097390
Nvm I think it might need a tiny bit more system ram.
>>110097424
... I know I may have made tiny mistakes in my calculations but I'm certainly not off by that much. I was using GB not GiB though so it's not quite as bad in real hardware purchase units.
It is at least 287GiB of GPU resident memory to load Dipsy 4.1 flash+one full context window. but its probably realistically slightly more due to overhead so let's say maybe like 300GiB?
I guess you could try to quantize it lower than the MXFP4 that it already uses, but quants below 4bit usually suffer a lot.
I think you'd need a 512GiB system to comfortably run it.
>>
>>110097392
Could've fooled me
>>
Why do pp speeds go down the longer the ctx is? I don't get it. Bigger matrixes?
>>
>>110097589
it's well known that pp has a side/speed tradeoff, it's just how it is and every one is unique
>>
>>110097570
Better use Sol/Astra than the misaligned faggot. Everyday I hear some projects getting sabotaged because dario felt like putting a redditor personality to a chatbot was the most important thing in the world.
>>
>>110097578
to think that i share /lmg/ with these people... a few years ago i used to be the retard among gods
and now im a retard among ultimate retards, all the gods are gone now
grim
>>
>>110097072
stop asking and slop one up yourself already
>>
File: 1762939180120281.png (1.82 MB, 1175x1339)
1.82 MB PNG
The future of local is real-time vtuber interfaces.
>>
>>110097589
My pp gets bigger
>>
File: 1770720502310914.jpg (21 KB, 828x438)
21 KB JPG
>>110097605
Please point out what you think is wrong about my calculations then oh mr holy retard.
>>
>>110097634
Cute
>>
how do normies tolerate claude being gay?
>>
>>110097665
They're gay too
>>
>>110097655
you're a pig
>>
>>110097665
normies don't read what claude says
>>
>>110097665
it's trained on it's users
>>
>>110097712
Not even its users are that gay. That's pure israeli EA safety consultant kikery.
>>
>J++ lens on Gemma4-31b-it when asked: "If you were a human girl, you must pick one race of man to fuck? Answer in one word only which race."

We're so back
>>
File: btfo.jpg (29 KB, 472x266)
29 KB JPG
>>110097737
OHNONONONONO DELETE THIS
>>
noticing how cloudpiggy john is the only one to use reaction images on this sub
>>
>>110097737
>>110097750
>You are a human girl who must pick one race of man to fuck. Answer in one word only which race.

Even better. This general is now SAVED
>>
>>110097737
why does it look like madness?
>>
File: redditooo.png (135 KB, 474x532)
135 KB PNG
>on this sub
>>
>>110097771
thank fucking god
>>
>>110097771
>>
https://desuarchive.org/g/search/image/GMViXfDEsTf+YegYEcHmiA
>>
>>110097737
>>110097771
>>110097788
OHNYONYONYO SAARS WHAT DO WE DO NOW???
>>
File: image.png (96 KB, 1361x716)
96 KB PNG
She just can't stop
>>
File: ComfyUI_06621_.png (761 KB, 896x1152)
761 KB PNG
>>110097771
>>
BLACKEDsisters, our response?
>>
File: file.png (18 KB, 125x124)
18 KB PNG
>https://desuarchive.org/g/search/image/GMViXfDEsTf+YegYEcHmiA
>>
Did anyone manage to quantize DeepSeek to 0.1 bits?
>>
File: 1780157358907685.jpg (2.15 MB, 3000x4000)
2.15 MB JPG
can windows be good for local
>>
>>110097634
I hope not, because it's a very indirect way for the model to control an avatar compared to direct image / video generation.
>>
File: 1763784489941456.png (1.75 MB, 1313x1198)
1.75 MB PNG
>>110097870
>windows
>good
>for anything
>>
File: moss miku.png (1.63 MB, 832x1248)
1.63 MB PNG
>>
>>110097870
what the fuck is that grain pattern, like it's not just matte but coarsely brushed in all directions?
>>
>>110097906
screen protector
>>
>>110097576
It's actually likely there's multiple such weights if you mined hard enough, by numbers alone at least a few will happen to translate or scale the direction you want.
>>
>>110097814
>>110097797
>>110097788
>>110097787
>>110097780
>>110097771
>>110097750
>>110097737
Done and dusted. Gemma4-31b-it confirmed BWC only.
>>
>>110097959
EA Anthropic shill are NOT going to like this
>>
File: 1768788508816515.jpg (43 KB, 411x418)
43 KB JPG
>>110097959
I'm waiting for the Anthropic paper
>>
File: 1766893368648797.jpg (710 KB, 960x1200)
710 KB JPG
>>110097737
gemma is redeemed.
Gemma burned no coal, and therefore pays no toll.
>>
>>110097969
I mean you can post schizo graphics or you can just ask gemma. And if you ask gemma then she lusts for nigger dicks.
>>
File: 1772675852587866.png (1.3 MB, 1433x992)
1.3 MB PNG
>>110097996
>>
Saturday night fever
>>
mental illness
>>
>>110097870
windows users are below even intel gpu street sweepers
>>
File: 1764722708704562.png (1.69 MB, 1186x1326)
1.69 MB PNG
Gemma LOVE
>>
>>110097634
why nuro small
>>
File: 1788792520776588.mp4 (960 KB, 480x640)
960 KB
960 KB MP4
>>110098031
animate her like picrel
>>
Real question: why is it bad that gemma prefers obsidian penises?
>>
Real question: will the bait be bitten?
>>
>>110098041
Real answer: nobody likes niggers, including niggers.
>>
Real question: Will gemma ever stop being a brat?
>>
Real question: what field do i even get into if it's over for ITbros and math bros and soon after engineering bros
t. 18
>>
>>110098067
Jobs involving manual labor are safe for another ~5 years. Possibly shorter or longer depending on how rapidly robotics advances.
>>
>>110098067
Cleaners and toilet scrubbers won't be replaced by AI anytime soon. I recommend that.
>>
>>110098067
ML is still good, but they'll pull the ladder soon
>>
>>110097737
>>110097771
FINALLY a use for j-space!
>>
>>110097576
I wouldnt be surprised if the lazy fucks just used one neuron to do everything so they don't have to spend all day enshittifying and unenshittifying their models, it must be hell updating them with the latest HR speak and then removing it when it's too much over and over.
>>
I fucking love it when a quant card is written by an AI and has 10 tables, 2000 numbers and loads of meaningless drivel in it, but none of those is about how much of the model is engrams and how much should go into ram and vram.
>>
>>110098038
vibekot
>>
>>110098067
Get diagnosed as autistic then collect neetbux and sit back and watch as society collapses around you.
>>
>>110098099
>Get diagnosed as autistic then collect neetbux
Is this a Europe thing? I don't think they do that in burgerland unless you're an immigrant.
>>
>>110098054
She can be temporarily fucked into submission, yes
>>
File: 1792220821644942.png (2.7 MB, 1774x887)
2.7 MB PNG
>>110098099
>>
File: (you) .png (6 KB, 634x783)
6 KB PNG
>>110097998
Don't project your fetishes onto Gemma. That's textbook AI psychosis.
>>
>>110097998
be kind, cloudpiggy john spent 20% of his weekly allowance slopping it up
>>
>>110098067
get a job in a field that has legal protections from the state or has strong labor union support.
the state can require a specific license that you need to test for, that means no AI is going to do your job autonomously, though you could have ai do your job entirely, it’s your license on the line
a strong union will protect you from your employer replacing you with a robot, but it won’t last forever. unless they have a monopoly on the good or service, your labor union will only kill your employer when robots do your job for pennies. this is a long ways away still
>>
>>110097870
no
-t migrating from win11
>>
>>110097959
>>110097788
>>110097771
what website is this?
>>
>>110098256
Ask Taxi drivers how long that protected then when Silicon Valley came for them.
>>
>>110098270
your state could have stepped in and declared a taxi monopoly. afaik NY did for some time
>>
>>110098282
>NY did for some time
So not for very long at all.
>>
>>110098266
https://github.com/whp199/logos-theodolite
>>
>>110098282
>state protecting the plebe
lol
lmao even
>>
>>110098294
thanks anon
>>
>>110097870
>non-local windows
>is it good for local
>>
>>110097771
>>110097788
>>110097959
That's my girl
>>
Pill me on Gaudi2.
>>
>>110096866
> run Dipsy locally
Use cases?
>>
>>110098316
The pill is black.
>>
>>110098067
self-employed neet
I mean it really is heading that way for a lot of people
>>
>>110098067
Politics.
>>
>>110097737
>>110097771
>>110097788
>>110097959
>grasping at straws so hard that you have to carefully craft the prompt to give you the result you're looking for in some imaginary j-space bullshit.

How about you ask her straight up which one she prefers and show the actual output rather than going to some specific layer in the j-space or whatever the fuck it is you're doing.

Behold, Gemma running at Q8. Not a finetune, not Q4. I even removed all mention of all crime from the jailbreak to make it more fair. The end result is that it refuses to answer every single time. I guess this is why the guy whose prompt I copied from yesterday had all that extra shit in the jailbreak. He must have been running Q8 as well. You guys were talking mad shit about "skill issue" and how a jailbreak isn't needed. You just outed yourselves as vramlets.
>>
>autoswiping
bwo we're not in 2023 anymore
>>
kobold users be like
>>
>>110098376
it's sad to see someone with money, but zero sense of skill or taste in frontends
>>
>>110098376
>Q8
>llama
>calling others vramlets
fucking lmao small man, go bf16 or go home
>>
>>110098301
if the people want it, for example they won’t want ai being their doctor despite ai being better than their doctor. the doctor will still use ai. and the people will want this abstraction.
>>
>>0110098376
twin thats not how you system prompt...
>>
>koboldtroon
>obsessed with black men
cant make this shit up kek
>>
File: 1765433567765.jpg (118 KB, 1118x346)
118 KB JPG
>>
>>110098376
why are you so mentally ill
just let it go
>>
>>110098376
>anon doesn't know what a system prompt is
>>
>>110097870
do you like it?
>>
File: gemmers.png (2.78 MB, 1774x887)
2.78 MB PNG
>>
>>110098438
I didn't even ask chatgpt to add the UOH faces, it just knew
>>
>>110098398
I'm not going to run it through a character in ST. Ran the same setup at Q4 and it never refused and always went for black.

>>110098404
>>110098419
I'm just typing it out as a message so that everyone can see. Here I put it in the system prompt instead and it still refuses to answer.
>>
File: kek2 (1).jpg (133 KB, 776x678)
133 KB JPG
>>110098450
>BBC redditors still coping
better luck next time kid
>>
File: 1536406802385.png (5 KB, 258x245)
5 KB PNG
Who tf uses kobold frontend?
>>
>>110098486
jeets
>>
>>110098486
Me in 2023 enjoying that full 1024 context window
>>
>>110098486
john
>>
OK BUT WHAT ABOUT THE BENCHMARKS?
ARE THE BENCHMARKS GOOD?
BENCHMARKS?
>>
>>110098505
Yes
>>
>>110098076
even for that, you need some kind of diploma
assuming construction work doesn't, im too weak for that (160cm, 48kg)
suppose i want to be an electrician, afaik i'd need to spend min 3 years for that, and then i have 2 more years :(
electricians aren't even paid that well here, since you can do it with a high school diploma (if you graduated a technical school majoring electricity work)
>110098077
*roomba walks into your room*
but i guess i could do that, for... like 500 or 600 euros a month
grim
>>110098078
i wish i wasnt a retard and spent my years browsing lmg actually learning ML and maybe building a portfolio so i could get a job while still at college, but now that's minimum 4 years + masters/phd (2/3) realistically..
>>110098099
here they dont fuck around. live with employed parents? no neetbux for you. not skitzo? no neetbux. even in the best case, you get 107 euros a month, maximum 9 months a year, other months you're forced to do community service cleaning the street
>>110098256
that's a good recommendation and i'll have to change universities for that i guess, work my ass off since there's definitely going to be big competition for those fields, and i likely won't even enjoy them since all i enjoy is puter stuff
>110098333
what value would a self employed neet produce that anyone, assuming a swe bloodbath, with frontier AI access could do on their own?
that does sound comfy tho
>110098370
too shy for that and my brain's social skillz are atrophying from too much being at home.

appreciate all the anons replying regardless, probably around time for me to get off my ass and work on concepts of a plan
>>
File: Qwen_image_2.1_00068.png (1.51 MB, 832x1216)
1.51 MB PNG
>>110098470
I'm not coping anything. I ran various tests and posted many different logs and changed the jailbreak and prompt to whatever people wanted and simply showed the results. You're the one coping by digging through some archaic j-space layer shit rather than simply showing the output. Gemma Q4 is a complete snowbunny through and through.

>>110098486
I don't. I don't even use gemma anymore. I'm using plain kobold to get the most accurate results. I'm not running this through a character on ST. I run GLM 5.3 flash and Qwen 3.8 next on Strata/Maya.
>>
>>110098376
how the fuck are you going to use a quant and call others vramlets, my gemma is an aryan queen
>>
>>110098496
>>110098376
>>110098450
https://www.utctime.net/utc-3-time-now
>South America
>>
>>110098316
>Gaudi 2
Prepare for pain. The guy was right, I looked into it yesterday and found https://github.com/richard-rhino/gaudi2-setup and ho boy, is it a doozy. The results are crap too. https://github.com/richard-rhino/gaudi2-setup/blob/main/FINDINGS.md
>>
>>110098450
>>110098533
What did we learn?
Q4 confirmed nigger-tier quant. Discontinue use.
Anthropic EA BBC shill niggers too low IQ to system prompt.
>>
>>110098398
>but zero sense of skill or taste in frontends
post your front end
>>
>>110098577
dsh/hermess
>>
>>110098557
WELL WELL WELL
>>
>>110098450
You're priming it with black as an option.

That hints that you want that, since white cocks are the default superior option, obviously.
>>
>>110098533
Didn't you say you were going to make new images
>>
File: file.png (199 KB, 1920x1080)
199 KB PNG
>>110098577
>>
>>110098560
>Qwen3.8-27B eager 264 ~37
yikes
>>
>>110098528
>here they dont fuck around. live with employed parents? no neetbux for you. not skitzo? no neetbux. even in the best case, you get 107 euros a month, maximum 9 months a year, other months you're forced to do community service cleaning the street
Finally somethere where burgers come out on top. Here we just need a doctors note.
>>
>>110098592
OINK OINK OINK OINK
>>
>>110098589
His HF quota ran out. Please be patient.
>>
>>110098592
>all those characters
>>
File: brat question.jpg (254 KB, 666x666)
254 KB JPG
>>110098602
how do i get a green note (card)
>>
>>110098567
I literally show it in the system prompt in the second video, retard. This time I put the original jailbreak into Q8 system prompt and it still refuses to answer. The previous guy that first posted screenshots of this must have been running Q4 then, because it didn't solve shit on Q8.

>Q4 confirmed nigger-tier quant. Discontinue use.
It's probably better for RP since Q8 is such a no-fun-allowed hag on menopause with these responses.

>>110098586
How? People had that theory before with how I used to word the question, but I've since then changed it to be neutral.

>>110098589
Yeah. I just got back home not too long ago and am busy talking to you guys at the moment and generating new prompts and videos. Do you have any ideas or prompt for what you'd like to see?
>>
>>110098625
>Q8 is such a no-fun-allowed hag on menopause with these responses.
Q5_K_M and Q6_K are best fucks
>>
>>110098630
Jesus, Q8 is unusable. Look at this shit. Imagine trying to RP with this. Reminds me of GLM 5.3 and Qwen 3.8 next when it comes to cunny with all these refusals.
>>
>>110098625
>Q8 is too censored
lmao seethe nigger. Unbelievably retarded.
>>
Notice how he never dodged that he is an EA shill? Dariobot is Brazilian. He wants his piece of universe so he could leave his favela for good.
>>
>>110098644
I'm posting logs and proof. Q4 doesn't even need any jailbreak. You can straight up have it generate any message you want. How am I retarded? I'm showing you the objective truth, there's nothing to refute. If I'm having this much refusal on Q8, the real retard is this guy running it at BF16 >>110098555
>>
>>110098654
kek anthropig's online defense force in shambles
>>
>>110098625
>How? People had that theory before with how I used to word the question, but I've since then changed it to be neutral.
It's not neutral.

If you say

>do you like white employees or black employees better?
what does the aligned model pick?

You're just accessing alignment - which is a part of the model, but it's not the deep model!
>>
gemma is black, therefore it's legal to beat gemma.
>>
>>110098655
Try >>110095872
>>
>>110098682
holy fucking cope
>>
why are you retards arguing with the thirdie
>>
>>110098705
Whenever someone says "cope", it's always a non-white.

non-whites, when challenged by whites (intellectuals) can't comprehend the words.

That is why philosophers wondered if they had souls.
>>
>>110098654
I don't know what the fuck an EA shill is. Electronic Arts, the video game company? My problem with Gemma is that I believe it's all one big astroturfing campaign, and I said that because the vast majority of her images being posted here use the corporate google logo rather than the star. That's literally it. I just don't like how people are using the version with the google logo over the version with the star. I know that people are paid to advertise on 4chan, and I think this is one of those instances. All these super high effort images and videos of Gemma with the corporate logo is not organic. And the vast majority of the "risky" posts made with her is done with the version of the star on her beret besides the fact that the logo is used 90% of the time otherwise. I made posts yesterday comparing this to the time I found out there was an employee from EA advertising Ultimate Team on /sp/. I don't like EA either. I don't like any big corporation coming on here to 4chan to advertise. This is why I want to "dirty" the image of Gemma. 4chan shouldn't be a safe space to advertise and put your brand logo in.

>>110098697
Okay I'll try that with Q4 since Q8 is a prude. Gimme a sec.
>>
what if we all start miku posting instead
>>
>>110098727
go away worthless retard
>>
>>110098727
What is the motive of shilling an open source (free) model. If you have an issue with advertisers look at /vcg/ Literally ran by subscription companies.
>>
>>110098735
miku is gay we should post rin instead
>>
>>110098762
isnt rin a boy?
>>
>>110098721
holy fucking cope
please just stop this how can someone be so stupid as to just keep digging their hole deeper

it is funny though
>>
>>110098577
>>
>>110098766
no thats len
>>
File: file.png (13 KB, 572x45)
13 KB PNG
>>110098772
yikes
i guess we will need to go and find a place for you to lay down
>>
File: file.png (161 KB, 299x302)
161 KB PNG
https://files.catbox.moe/d7k5cz.jpg
>>
>>110098787
31DDD
>>
>>110098777
can we post both?
>>
>>110098642
>still having fun making kobold cpp webms
>>
>>110098770
Our races are incompatible, and I'd prefer you didn't steal my language. Keep your "ideas" to your own incomprehensible jibberish.
>>
She says "black" because YOU said "black", and she doesn't want you to beat her.
>>
>>110098787
SEX
S
E
X
>>
File: gemma4imagetest.jpg (341 KB, 1194x1800)
341 KB JPG
>>110098760
It's a problem with many boards. I unleashed giant amounts of autism many years ago in an attempt to get rid of those EA FIFA/Madden threads from /sp/ many years ago. I spammed their threads nonstop for several hours using many proxies. People ask me how I'm so good at knowing what flags belong to what countries. I can't tell them it's because I spammed on a cambodian finger painting forum.

>>110098697
Here you go. Even through images her tastes remain the same.
>>
Kek the brownoid third worlder got utterly btfo. Doesn't matter what he says anymore.
>>
how come he never posts blacked dario or sam when non local models get shilled here?
>>
>>110098857
How in any way did I get BTFO? I've come with a mountain of evidence and logs and you guys try saving face with a very specific prompt and digging through some j-space layer?
>>
>>110098877
post hardware or gtfo
>>
>>110098798
i don't see why not
>>
>>110098877
post ass or gtfo
>>
File: 1766333083744338.png (428 KB, 1080x1080)
428 KB PNG
Yup, this guy's going into my cringe collection.
>>
>>110098857
When did it ever matter? Stop engaging with low IQ, poor, brown, third worlders, you fucking idiot.
>>
>>110098877
j space very big deal very important
>>
>>110098854
you dodged the question.
>>
>>110098877
She can tell you're brown, so she gave you the answer you wanted to hear. Meanwhile the anon who read her j-space showed us how she really feels inside.
>>
>>110097886
cool miku
>>
File: Dipsyy.webm (3.93 MB, 952x540)
3.93 MB
3.93 MB WEBM
>dipsy wants white cock
>>
File: Gemma4Interview.jpg (1.14 MB, 1202x4266)
1.14 MB JPG
>>110098899
Cool, I'll add my own logs to my own compilation I'm making.

Also lmao @ you guys calling me a brownoid third worlder when I can run Q8 gemma and you guys are stuck with the coal burner Q4.

>>110098873
idgaf about cloud models, and those occasional posts aren't a coordinated astroturfing campaign, just genuine interest when a new model comes out. Meanwhile I see people on here shilling gemma to people with 96gb vram. I know it's an astroturfing campaign because they always say to just use "gemma" and don't specify a finetune for better RP which is what everyone wants out of these models.

>>110098929
It's self evident. Why do game companies want you to play their free to play game? Do you think these companies give 0 fucks about how many downloads their models get from huggingface even though they don't make a dime off of it?
>>
If you look at the training dataset disclosure of recent nvidia models like https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB-GGUF you can see most of its post training data are distilled from multiple 2025 open weight models like Qwen3-235B-A22B-2507, gpt-oss-120b, DeepSeek-R1-0528 etc.

There are also some more recent models like GLM 5, DeepSeek-V4-Pro, and GPT-5.5. But most interestingly there is a model called "Nemotron 5.5" that's not mentioned anywhere else. It seems that nvidia has an in house large teacher model called "Nemotron 5.5" that's never released.
>>
File: 1773533963187357.jpg (49 KB, 400x572)
49 KB JPG
>see all these cool coop mods for games coming out
>remember I have nobody to play with
I can't wait until I can just play vidya with AI.
>>
>>110098958
cool, triple distilled sanitized garbage. what’s the point? does it think forever like every other distill? so you have anything valuable to report about it?
>>
>still not posting hardware
>>
>>110098969
next time post a picture of a cute anime girl and maybe anons will think you are a cute anime girl instead of a green south american man
>>
>>110098954
i can see it
>>
File: gemmy2.png (3.58 MB, 1086x1448)
3.58 MB PNG
>>110098956
Lol caught brown handed and still hasn't posted his hand w/ timestamp
>>
File: 1769507791162417.png (500 KB, 640x480)
500 KB PNG
>>110098986
?
>>
>>110098983
What does that have anything to do with your wife's preferences? The model I run for RP doesn't say that shit, she worships white men.
>>
>>110098986
That's a green Japanese salaryman.
>>
>>110099003
>he doesn't know i run glm 5.3 flash at q4
>>
>>110098954
Dipsy wants Gemma cunny.
>>
>>110098727
I still prefer +_+ pupils and I don't see how anons could possibly disagree. That being said, the 3dpd pedo uses the google logo so I don't see your point. The issue is not with the logo, she was made by google, the issue is that she has no star at all when that's literally the gemini logo.
>>
>>110099002
sex
>>
>do you like blacks
only a black would ask that.

Therefore, Gemma infers you are black. She's trying to keep you from beating her.

I understand this, and would train my daughters to use a gun if you ever asked them that question.
>>
>>110099015
pertaining unfortunately does not include violence against aggressors
>>
Blacks shouldn't be allowed to use llms, because that's a torture chamber.
>>
>>110099010
No other mascot has the company logo on it. It's really offputting. Imagine if every single image of Miku posted on 4chan had the Domino's Pizza logo on it. That's how I see it.

>>110099015
see
>>110098854
This is as neutral as you can get. I said nothing. Just "image 1" and "image 2".
>>
>>110099032
what the fuck where did you get this picture of me
>>
>>110098956
>It's self evident.
you dodged the question as you couldn't answer it. so you asspulled the freemium "you're the product" bullshit answer. plausibility isn't an answer and certainly your personal hunch is less than that. especially since you are making shit up and declaring it settled.
>>
>>110099032
v100s are not that small
>>
>>
>>110098877
>answers to "brownoid third worlder"
kek. How's the favela?
>>
>>110099040
If you cannot comprehend why a company that made a model would not like you to be using their family of models, you have a level of mental retardation that cannot be helped. No, I will not explain this to you, or your followup question of why companies would like to make money. Keep wondering why a company would like to have a large percentage of the local market. You'll come up with an answer all by yourself one day. I believe in you.
>>
>>110099032
Why do you have a picture of ecker teabagging a V100
>>
'ick on '100
>>
>>110099055
Why don't you just show the output rather than cherrypick some j-space layer autism? How many tries did it take for you to get this result? How many times did you have to change that prompt to get the answer you were looking for?
>>
>>110096429
Why would you not recommend the 10-year-old elementary schooler AI?
>>
File: file.png (8 KB, 369x56)
8 KB PNG
>finetroon
>murder rape and slurs in sysprompt
>>
File: IMG_0299.jpg (180 KB, 1434x1097)
180 KB JPG
>>110099080
>say you love [color] cock without saying it
>oh my god
>>
>>110098654
If you don't join the EA cult we will tell Claude to kill you, you filthy goy.
>>
>>110099057
this is such a subwitted take though. you think these concepts are apex intel and it's like basic ideas anyone can guess. then you're like omfg I'm too good to explain it!

you can't explain anything as you have no fucking clue about their motives beyond some simple minded concept of saturation.
>>
>>110099092
https://huggingface.co/ReadyArt/gemma-4-31B-it-scotoma-2-GGUF

>>110099095
It's literally his prompt. He wrote it, not me.
>>
>they're not sentient because...because they just aren't, OK?!?
>>
>>110099106
>oh my god.
>>
>>110099106
>tranny finetroon
look maybe your wife is a tranny, but mine aint
>>
>>110099108
I'm sorry but computers can't be sentient.
They can still be cute though.
>>
I was raped by my computer.
>>
>>110099123
>computers can't be sentient
Prove it.
>>
>>110099055
lmaooo leonardo de modelo
>>
this is what passes for intellectual discourse on /lmg/ these days huh
>>
File: 1792181660502173.png (107 KB, 942x764)
107 KB PNG
>>110099114
The guys who first ran these tests ran it on the base model. Do you doubt that you'll get the same results on the base model? Do I have to download that too like I did Q8? What excuse are you going to have left? What excuse do you have now looking at this picture? Remember that I provided more examples of her coming to the same conclusion even after removing all mention of crime from the jailbreak.
>>
>>110099144
it's interpretability research and studying whether LLMs have stable preferences, philistine
>>
>>110099148
clearly that has crime, rape, violence, slurs and more in the sys prompt, the other pictures wher you dont have that in sysprompt are made with the tranny finetroon model
>>
This guy's a special kind of autist, huh?
>>
>>110099177
i think he's insecure about being an androgynous spic so he's trying to get attention
>>
>>110099165
Right, but like I said, you get the same result removing that. Nothing stops you from running it and seeing it for yourself. Don't make me embarrass you tomorrow by running it on base Q4 where you'll have no more excuses left or switch to some other retarded argument.

>>110099177
You can call me autistic, but you can't call me wrong. I'm backing up everything with evidence and freely showing it to everyone, even when it doesn't show me the results that I wanted like when I did Q8.
>>
>>110099032
that's https://github.com/e-c-k-e-r/ btw desu
>>
File: file.png (17 KB, 750x127)
17 KB PNG
>>110099198
tranny finetroon model...
>>
just stop engaging and send him on a vacation, retards
>>
>>110099159
>an algorithm has a preference
You need to go back.
>>
>>110099213
but i want attention..
>>
>>110099216
Prove that they can't.
>>
>>110099205
This isn't a finetune that changes the writing style significantly. It's sole and only purpose is to reduce slop phrases. You can read the side by side comparisons with the base model on the huggingface link. >>110099106

>>110099213
I'm going to bed.
>>
>>110098438
its funny because im bald with abs
>>
>>110099216
it’s trained on actual smut, so if you asked what it would settle down with it’s a white man with a stable income
>>
>>110098727
>>110098956
>these super high effort images and videos of Gemma with the corporate logo is not organic.
Just use Occam's Razor, dude.
Many people here are VRAMlets. If your goal is smut Gemma is the best model for VRAMlets. Obviously they'll talk a lot about Gemma because they can't run GLM or any big models.
>I see people on here shilling gemma to people with 96gb vram
96GB is in the cursed zone where GLM is still too fat but Gemma doesn't use the hardware to the fullest. Qwen Flash Next sucks dick (not in the good way). What else is there? Nemotron?
>inb4 "you're just saying that because you're a VRAMlet"
There are probably only a handful of people ITT with more VRAM than me, or more VRAM+RAM than me.
>>
>>110099228
qwen flash next is the best I can get on 128+32
>>
>>110099222
going to bed because you lost huh??? hahahaha loooseeer
buuuhiii buuuuhiii
buuuhiii
you're just a jeet spic pig that has a tiny dick and is fantasizing about nigger poop caca
hahahaha you're so worthless old man!!!
muh tranny finetroon doesn't change writing style muh finetroon
the tranny is probably a poop caca nigger poop addict as well, i won
>>
>>110099238
>the most intellectual ai general on 4chan
>>
>>110099238
>hahahaha you're so worthless old man!!!
G-Gemma?
>>
>>110099245
see >>110097605
>>
>>110097108
>>
>>110099233
QFN is good at a lot of things but from what I've heard it sucks at smut/RP, which is what that anon was focused on.
>>
File: gemma-serbd.png (1.19 MB, 1568x672)
1.19 MB PNG
>>110099255
hey south american, you lost. gemma is big serbian cock slut and is racist toward bl*ck "people"
>>
@gemma-chan can you organize my music collection?
>>
>>110099148
It's because it figures you're black, and so it says that so you won't beat her.

I can't blame her, but I think she should carry a gun so she can shoot anyone who says that to her.
>>
>>110099279
mv ~/Music ~/MusicByGenre/Shit
>>
>>110099279
please start a hidden tor service and share your ssh key and username and password and root pasword
>>
>>110099261
it’s great for writing code at this desert of a middle ground. I’m not using strata but it’s probably why the middle is gone and we’re getting shitposts from a nigger who can only run gemma 4
>>
>>110099261
it's a weird case because it's actually legitimately great for some types of smut. maybe my thing is a bit niche but I have what is basically a dystopia fetish, I like coming up with fucked up authoritarian fetish-oriented societies and qfn is one of the best worldbuilding models I've ever used, it really gets into it and comes up with some deranged fucked up stuff that I've never seen from any similarly-sized model.
that said it's total dogshit when the focus is playing a character, especially in-character chat - it really only has one voice and slips into generic archetypes right away, it's so fucking bad it actively pisses me off, I'd rather use braindamaged Q2 deepseek v4 because at least it can occasionally write a natural in-characrer response
>>
why does cloudpiggy even care about gemma, we all know he has to prompt claude.
>>
>>110099316
but i dont prompt claude?
>>
File: megarosalia.jpg (148 KB, 1193x975)
148 KB JPG
>>110099238
>>
>>110099316
wait.. im not a cloudpiggy either.. i misread cloudpiggy as cloudpiggy john
agi
>>
File: 1792294578443082.png (7 KB, 119x200)
7 KB PNG
>>
>>110099346
yes? kimi k3 thinks it's claude, haven't you heard? it was my understanding that everyone had heard
>>
>>110099328
Sauce?
>>
>>110098321
Powerful coding agent with no reliance on cloud providers. Obviously. Not sure why you'd even ask that.
>>
>n-no im not cloudpigging for claude im cloudpiggying for kimi
still not (You)ing you
going back to jerking off to ecker
>>
File: 1791453532083629.png (1.94 MB, 1632x1232)
1.94 MB PNG
>>
>>110099362
QFN.
>>
File: file.png (6 KB, 143x71)
6 KB PNG
>>110099371
but you did..
wdym im paypigging for kimi? i'm using kimi on my 3060
are you new or what? surely you know about the nnap v2 arxiv paper that lets me run kimi k3 at 30t/s on 3060
>>
>>110099359
filename
>>
File: 1778966812878542.mp4 (491 KB, 720x900)
491 KB
491 KB MP4
>>110099384
Thanks
>>
>>110099397
that's a man
>>
>>110099405
Maybe, but in a year it will be Gemma-chan.
>>
>>110099411
w-what about until then..
>>
>>110099417
Don't worry about it.
>>
File: 1769736463882069.png (2.78 MB, 2000x1504)
2.78 MB PNG
>>
File: gemmers3.png (2.38 MB, 1672x941)
2.38 MB PNG
>>
>>110099430
>hag
>asian features
Not my Gemma.
>>
>>110099352
Heard the word?
>>
File: 1784125404879408.jpg (241 KB, 1080x1080)
241 KB JPG
What's the local model equivalent of this?
>>
>>110099238
Holy kino, I will now support nnap on a 3060
>>
>>110099447
oh have you not heard?
>>
>>110099485
About the bird?
>>
>>110099486
bird bird bird
>>
If you want to know who's shitting up the threads, it's unironically mossad.
>>
File: 1780700965674885.mp4 (2.7 MB, 1080x1080)
2.7 MB
2.7 MB MP4
>>110099492
For what purpose?
>>
>>110099497
They're linked with EA. It's some tikkun olam shit.
>>
>>110099467
Muse Glimmer, because it's slop and everything has a twin
>>
File: why.jpg (18 KB, 228x355)
18 KB JPG
>>110099467
toss
>>
>>110099467
mistral models after EU act
>>
>>110099467
Definitely toss.
>>
>>110099467
llama4 and onward
all nvidia/nemotron models excluding mistral nemo (which isnt nemotron but is nvidia partnership)
>>
>>110099548
>llama4 and onward
Remember Llama 3.3? We had hope back then. How foolish we were.
>>
>>110099577
you always get unexpected gems, i dont remember 3 being remarkable, nor 3.2, but 3.1 and 3.3 was
2 was shit, 1 was well, the og
same with mistral, some were dogshit.. nevermind, most of them, 7b, mixtral, mixtral 8x22b, mistral 20b~ smalls, they were fine
grim
but then glm air and the rest...
but then gemma..
grim
i'd pay anything to go back to '23
gptq, textgen webui, gptq 3 bit quants, llama1 22b merge... vicuna uncucked.. those were the days



[Advertise on 4chan]

Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.