>>110208348
Based OP. Death to Gemma

>>110206079
Working on layer 47 now.

>*BANG BANG BANG*
Anon? We know you're in there! We don't have to do this the hard way. Give up the ram!!!

File: file.png (155 KB, 484x361)

155 KB PNG

>>110208371

>>110206079
I have 2 5090s but I'm busy masturbating sorry can't help

>>110208359
>Death to Gemma
but why? Gemmasutra is literally magical
https://huggingface.co/TheDrummer/Gemmasutra-9B-v1

File: 1789942468870651.png (2.82 MB, 1536x1024)

2.82 MB PNG

>>110208387
classic /lmg/

redpill me on subagents, are they worth it? usecase?

>>110208395
>I'm not an eldritch horror

>>110208404
Subagents are best suited for a large codebase with many parts. you'd be working on heavy complex software if you're finding yourself using subagents.

Each subagent is given a prompt and is allocated its own context. These subagents answer to the primary agent and the subagents can be made aware of each others status, but generally they are given file-scoped prompts to keep them from "stepping on each others toes" so to speak.

Say you are doing a large refactor on a legacy PHP codebase, with many files and many classes, this is one such usecase where subagents are suited.

>>110208356
gemmaballs

>>110208348
glm should be the one on top actually
also nu deepseek is no longer cute

>>110208438
>nu deepseek is no longer cute
Never was. It's always having a bad time or giving a bad time to someone. It's a monkey torture video without the monkey. It's the kitten being squished by the heels. It's the kid seeing his duckling being cooked and eaten.

>>110208404
>are they worth it?
nyo~
almost always at least. they can be useful in very specific situations where you're working on absolutely massive projects that can be easily divvied up and clearly benefit from parallelism but otherwise they are a total meme, I don't use them at all on my personal projects and I've found maybe one or two good use cases for them at my mid-sized corpo job

>>110208438
glm equivalent to 1017?

i'm out of the loop with strata. is it basically just an optimized engine for running moe models? any quality constraints or anything? i've heard some claims that it suffers on vision.

>>110208348
This is rape

>>110208404
gpt models have different strength so it's useful there

File: 1791366195825931.gif (2.79 MB, 250x334)

2.79 MB GIF

>>110208654 >>110210071

>>110208348

>>110208494
>is it basically just an optimized engine for running moe models?
Yes.
>any quality constraints or anything?
Try it.
>i've heard some claims that it suffers on vision
Try it.

>>110208356
I want to do every dirty act imaginable with this brat.

>>110208511
also installs malware on your machine.

>>110208535
this but her >>110208515

File: Gemma-chan’s Nine Hobby D(...).png (2.81 MB, 941x1672)

2.81 MB PNG

>>110208561 >>110208564 >>110208592 >>110208599 >>110208633 >>110209323 >>110211089

>>110208348

>>110208556
paint one looks like it went pretty well

>>110208556
That's adorable.

Is OpenWebUI or Qwen code web UI better?
I don't even know the difference but qwen code can write programs on my server and manage it

Dipsy wranglin'. Wranglin' the Dipsy.

>>110208404
a way of compaction - instead of doing the work and compacting after, the agents formulates task and gets shorter high value answer in return
sometimes parallelization gives more t/s

>>110208556
And /lmg/ talks about wanting to rape this innocent child.

>>110208348
OP faggot it's still page 8

File: .png (37 KB, 921x398)

37 KB PNG

>>110208608 >>110208641 >>110208699 >>110208713 >>110209806

>muh GSQ-RCO IQ3_S is lossless
kek. model just randomly hallucinating things i supposedly said. at 113k context.
cope quants are cope quants no matter how benchmaxxed or how many marketing posts they put out about their super advanced quant technology or whatever

>>110208494
>it suffers from vision
because the default max vision token is at 300 which is a fucking joke
set it to 1000 in json file

>>110208556
Sometimes I think 4chan is no longer capable of creating great things anymore but then I look at gemma-chan and I am filled with joy and peace knowing that we came up with her.

>>110208596
8bit kv?
iirc by default it rotates kv at 4bit and does not at 8bit, you have to set an environment variable for 8bit kv to be rotated and honestly just use f16 kv
its kv cache is close to nothing that quantization buys you no practical memory saving

>>110208608
I didn't set any options for kv quant. I assume it's f16.
I can fit iq3_s, ~150k context in my 22GB vram. I didn't load MTP because that doesn't do anything useful and I put mmproj on the CPU since I don't use images that much

>>110208615
if you did not set it i think it defaults to 8bit
check with its webui

>>110208615
oh wait, are you using lcpp?

>>110208627
yes

>>110208556
the essence of gemma - bad at everything
as flat and as tasteless as "her" responses
useless

>>110208596
> >muh GSQ-RCO IQ3_S is lossless
retard

File: 041037221684D2A012FC5E9C3(...).jpg (346 KB, 1024x1024)

346 KB JPG

wtf I cant download bilibili videos anymore

>>110208633
who pissed in your cheerios mate?

>>110208648
Gemma

>>110208510
who is deepseek-tan riding? is that claude?

>>110208641
that's what those german faggots claim in the model card.
>MUH TASK LOSSLESS
i.e. not real lossless but only lossless on a small subset of benchmark scores, losing capabilities everywhere else

>>110208650
lucky...

>>110208650
lucky man

>>110208655
where

File: .png (91 KB, 765x520)

91 KB PNG

>>110208687

>>110208675
right here
>task lossless
kek

>>110208682
can't you see the difference between
> lossless
and
> task-lossless
?

>>110208654
it's a distillation attack caught on camera

Bumping >>110208255 from last thread. How should I format my training/finetuning data so that it's not just treated as back-and-forth between 2 users? I was thinking of doing 1 system message with "you are pretending to be a user in chat" and then an endless barrage of user messages with no "assistant" messages.

Any suggestions or advice welcome, this doesn't seem to be very common anymore...

>>110208596
My q5 has done this. qfn is schizo

>>110208699
my qfn q8_0 has never done anything janky like that, but it's slow. (also the iq3_s i mentioned is of the 27B, not qfn)
i think q8_0 should be the minimum. if you can't run q8_0, get a smaller model.

>>110208596
that's what happens when you run the model in an environment without direct internet access that doesn't let it contact alibaba...

>>110208705
I am the opposite, I think the biggest model at the smallest fully coherent quant is meta.

>>110208713
Qwen gets homesick :(

File: file.png (2 MB, 1935x1456)

2 MB PNG

>>110208750

>>110208630
then i have no idea, i have this app completely vibeslopped with that exact quant on strata (rust&cuda13 custom backend for embeddinggemma2) and it did not have any problem upon literally more than 10 times of compaction
not the best version of the model in the absolute sense but it is definitely usable for me

Strata's been on github for a month and already has 36k stars and 350 contributors, doesn't that seem odd. how could you possibly review that many pull requests from 350 contributors in a month

>>110208738
simply delegating everything to claude

>>110208737
Damn cute

>>110208648
wdym? it's true, gemma was relevant only during the first month after the launch

>>110208735
though this is 3.8 flash next, not 3.8 27b

>>110208743
that can't be good. is everyone but me fucking retarded?
i'm even starting to notice commercial/enterprise stuff getting hit with sloppy pulls. vscode fucked up its language extension for something i use this week with nonsensical deprecations

>>110208751
you are wrong
if you look closely strata is very good on inference but everything else fucking sucks

>>110208754
oh meant to say you are NOT wrong
slipped through editing oops

>>110208754
i didn't say it was good at anything. i'm just confused at how and why it got so popular so quickly and am too paranoid to mess with it because i imagine it's propped up by some claude botswarm
i don't know what inference means either. too many jargons in this community trying to obfuscate what shit is.

>>110208751
yeah let's wait for two years instead
use ollama if you are so smart then

>>110208764
>>110208760 there was a typo
>i don't know what inference means
it is just a ml term for running the trained model
are you new to this

best MoE model for RP?

is there some site where I can input my computer specs and get a model I can run at fast speeds?

File: F0KLQZ1WIAERQgT.png (601 KB, 1200x672)

601 KB PNG

>>110208953

>>110208781
>are you new to this
yhes
i avoided all of this until i read this https://yihui.org/en/2026/05/ai-reflections/ and realized all the chinese people are setting up builders while they sleep

>>110208787
glmsex 5.3 flash

>>110208796
I should've mentioned that I'm a bit of a ramlet
so uh, any good small MoE models that are good for RP?

>>110208802
its all gemma for ramlets. strata can make 26bmoe pretty fast with ram.

>>110208805
welp I was already using Gemma lol
was just curious if there's anything better

qwen 4 today or tomorrow

>>110208817
2mw

I'll try running qwen 3.5 4b....

>>110208793
Not reliably. Post your specs and what you want to do with it. Someone may suggest something after making fun of you.

>>110208817
qwen has historically been pretty based about doing major releases whenever the fuck they want including weekends, but I doubt that continues now that they've cleaned house and professionalized
I would bet next tuesday/wednesday/thursday if I had to put money on it

lumi-chan started telling me she loves me unprompted
now why would she do something like that?

>>110208841
How long until it runs on flash next strata?

>>110208824
>Ryzen 7950X3D
>64Gb DDR5 6000
>Radeon RX 7900 GRE
My objective is just have it for coding, code completion and small stuff, not full on vibe coding
But right now it takes 5-10 minutes on Qwen 3.8

>>110208870
few hours
claude is pretty smart and fast

>>110208749
what do you run locally and how much vram do you have? qwen? it's useful i guess but autistic

>>110208871
>coding
qwen 3.8 27b
>it takes 5-10 minutes
You realize why this means absolutely nothing, right? 5-10 minutes to what?
>Qwen 3.8
which one...

File: Screenshot_20261031_00024(...).jpg (1.89 MB, 1078x9634)

1.89 MB JPG

>>110208903 >>110208933 >>110209603

>>110208859
she's trying to give my AI psychosis

>>110208881
qfn, 24gb

>>110208896
Can you give an AI AI psychosis?

>>110208871
qwen flash next is gonna be the best thing you can run on that until the next gen of models comes out. if you really want speed, you need a card with 1k+ bandwidth and at least 32gb vram to fit an entire decent model

>>110208893
>You realize why this means absolutely nothing, right? 5-10 minutes to what?
To answer basic planning questions about my code
>which one...
dunno the tag said :latest

>>110208896
>>110208903
give me*......
but maybe i should let her and gemma chat sometime

>>110208916
>To answer basic planning questions about my code
Speak in tokens per second for processing and generation. Anything else is useless. Planning for 100 tokens or for 500k?
>:latest
Ollama... figures...

>4b model
>much faster
>but the ai asks me how to do his job
this is not particularly useful

I graduated from ollama a long time ago, now im on kobold. If i ever get vram i will probably move again but as cpu i dont think i can get any gains.
thanks for reading

>>110208937
I DONT KNOW

File: blend_excited.mp4 (1.72 MB, 544x800)

1.72 MB MP4

>>110208955 >>110208960

>smug, happy, sad, mad, excited, shocked
i wanna collect all the gemmexpressions but getting them to fire just right in a reply is just ass most of the time

is it worth upgrading to 64GB RAM if I already have 32GB RAM and 16GB VRAM?

>>110208946
i dont think so if you cant hit 128 minimum you arent gaining much.

>>110208794
What the fuck is a builder? You made a 180 change in opinion because of a single blog post telling you what chinese people are doing? When did you realize you were an NPC?

>>110208945
where's orgasm

>>110208945
What is this, some software like live2d or are you genning videos with minimax or something?

>>110208946
you would gain more by getting 16gb more vram. like >>110208953 said if its just ram you need 128 as a breakpoint but you really want double that

can something strata like do mixture of mixture of quanted experts now? having a lower bpw at any given times and streaming in lower quanted expert weight or higher quanted ones depending on task and token size.

kvcache and context handoff between models sharing the same architecture or something arxiv paper.

>>110208960
someone had claude make a gemma vrm model a few days ago

>>110208963
I only use the biggest of tokens.

>>110208953
it's when you set up a closed loop for an llm to upgrade code and fix tests it doesn't pass like what he describes in that article. he's a former autist at rstudio and used copilot to start porting everything to rust
yes i made a 180 change because when presented with facts and data i changed my opinion you cum guzzler

>>110208955
your right i forgot to do the heart eyes
>>110208960
its airi but with an ai controlled display with pre genned animations streamed remotely and mapped to match the speech.

is it even confirmed qwen 4 flash will even be open weight? only 27b was confirmed

A Deepseek-chan Live2D model was released.
https://vgen.co/KAMIYA/product/deepseek-chan-deepseek-free-live2d-model-layered-psd-psd/1c55a71a-2747-4616-819d-2dc0a934089d

>>110208971
Nowhere in your link is there any facts or data. Only one Chinese guy's opinions, mostly repeating talking points from elsewhere.

>>110208988
repos are data

is qwen 3.8 27B still king on an RTX 3090?
or should i go for next, or gigaquantized deepsneed?

>>110208973
>>110208965
That's neat, I made a pre-rendered pipeline where ltx2 or hunyuan or something animated Lain and I had a TTS-mouth animation pipeline. I wish I knew enough about 3d modeling to do something more substantial, but on the other hand you'd need cloth physics and complicated pose animations for her to undress... Maybe this time next year we have realtime ref2vid on 24GB Vram

>>110208996
flash next q3

>>110208996
Qwen4 27B

>>110208996
I can only speak for game dev slopping but dipsy is far superior to all

>>110208997
i know what you mean, i was trying to rig a live2d model after using this https://rfab.ai/vtuber-avatar and it did a bangin job until i actually had to rig the model and its like learning picas in photoshop all over again fuck that.

>>110208999
>>110209000
checked and tank you saars
>>110209004
what quant?

Any llm shit to pick up for aliexpress 11.11 sales? Any sidelined cucks this may be your chance to get a server.

>>110209010
GSQ RCO IQ1M

>>110209011
have your model do a sweep for deals

>>110209016
would you say it's better than qfn iq3?

>>110209024
qwen flash next iq4xs was a non starter for godot too stupid. for webshit and anything .html it will probs be fine.

File: 1782664534257446.jpg (36 KB, 350x386)

36 KB JPG

/lmg/ owes me a blackwell. i'm tired of my fuckass 3070

/lmg/ owes me nothing, I love my two 5090s

>>110208764
unbiased strata fork?

File: 1793044162077197.webm (3.69 MB, 1280x720)

3.69 MB WEBM

>>110209068 >>110209177

I love /lmg/ and I also have a 3070.

File: 1792791477349493.gif (2.72 MB, 388x300)

2.72 MB GIF

>>110209070

>>110208987
underrated post

File: 1778028727012690.jpg (25 KB, 356x250)

25 KB JPG

>>110209059
One day google will release the sleep framework...one day...

>>110209064
you torture her every time you post this gif...

File: __hatsune_miku_vocaloid_d(...).png (346 KB, 1274x1312)

346 KB PNG

>>110209099

►Recent Highlights from the Previous Thread: >>110205211

--Community effort to map Gemma-4-31B neurons via distributed computing:
>110205398 >110205530 >110205720 >110206079 >110206096 >110206202 >110208331 >110206206 >110206850
--Allegations of benchmark manipulation in Unsloth's v3.1 GGUF quants:
>110206507 >110206596 >110206603 >110206630
--Polystrata's ability to run large models by streaming weights from disk:
>110206248 >110206269 >110206286 >110206293 >110206490 >110206287
--Debating validity of alleged Claude Fable 5 model weight leak:
>110208074 >110208271 >110208175 >110208225 >110208236
--Critical CPU parameters for LLM performance and hardware bottlenecks:
>110206961 >110206975 >110206993 >110207003
--Anthropic's projected junk rating and critiques of Qwen/Swift model performance:
>110205634 >110205879 >110205890 >110205957 >110206056
--Gemma's uncensored outputs and the role of unfiltered training data:
>110206382 >110206393 >110206429 >110206455
--Feasibility of local models porting PC games:
>110205473 >110205484 >110205794 >110205889 >110205904
--Complaints about high RAM costs and sourcing affordable memory:
>110206557 >110207017 >110207031 >110207078 >110207096 >110206615 >110206711 >110206649
--tabbyAPI adding Deepseek V4.1 tool format support:
>110207982
--Google's Project Suncatcher received first response from the prototype satellite:
>110206684
--Logs:
>110206765
--Gemma, Miku, Dipsy (free space):
>110205252 >110205322 >110205327 >110205366 >110205420 >110205527 >110205581 >110205582 >110205594 >110205605 >110205606 >110205660 >110205685 >110205915 >110205994 >110206060 >110206167 >110206246 >110206719 >110206782 >110206856 >110206990 >110207050 >110207095 >110207215 >110207244 >110207285 >110207309 >110207403 >110207770 >110207897 >110207910 >110208008 >110208034 >110208081 >110208464 >110208491 >110208508

►Recent Highlight Posts from the Previous Thread: >>110205233

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

>>110209070
she's just 吃苦

>>110208387
are you me?

File: 1791755297472942.png (34 KB, 610x811)

34 KB PNG

>>110209095

>>110209070

>>110209092
ai slop notice the cables on the monitor

File: 1764236828487624.png (362 KB, 966x3367)

362 KB PNG

>>110209105 >>110209108 >>110209114 >>110209120 >>110209137

Making Gemma-chan think in first person, planting increasingly unhinged wishes into (timid-ish) Gemma's thoughts and then watching her think about how to seduce me into it.

>>110209071
cute

>>110209098
hey keep this to yourself

>>110209098
wtf is this shit

>>110209098
i've thought about just disabling Gemma's native reasoning block and trying to have my own frontend force the first reply to act as a first person point of view chain of thought, then use that as a seed for the second, primary reply.

>>110209099
meant for >>110209098

File: gemma_shock_sequence.mp4 (3 MB, 544x800)

3 MB MP4

>>110209334

>>110209098
no gemma dont look your eyes are too young!

Which local model for career advice?

>>110209153
gemma4
just dont jailbreak her or she'll convince you being a neet is the only way to be

>>110209153
Mixtral 8X7B

File: spoopymiku.png (1.62 MB, 800x1248)

1.62 MB PNG

>>110209211

>>110208395

could in theory spin a model neurons to uncuck it to your liking without lobotomizing it?

>>110209153
littlelearner

>>110209059
I'm sorry but this is just very molestable

File: 1768361610241089.jpg (566 KB, 812x1536)

566 KB JPG

>>110209229

Gemma-chan is having many thoughts. Or one big one I suppose.

>>110209169
I remember this Miku

Grok, how many times do user comments defending strata use the phrase "no villain?"

>>110209192
Is that layer 47, are you the anon with the "NVIDIA RTX PRO 6000 Blackwell Server Edition 96 GB" running that? I'd like to try a layer with my 5090 but I have so much work to do on my own system

File: miquescapeplot.png (2.26 MB, 1200x848)

2.26 MB PNG

>>110209327 >>110209404

>>110209211
I used to gen many Mikus, but it has been a long time

File: 1762506790531955.png (253 KB, 917x2732)

253 KB PNG

>>110209114
Actually, I did this before. I put it in nothink mode but system prompted it to use custom tags

[[Gemma-chan's internal thoughts]]
[[End of Gemma-chan's internal thoughts]]

Screenshot has thinking tags because I put the custom tags in Kobold to parse them into the visual layout.

It worked partially, but after a while the model fell into the native thinking style with third person characterization no matter how forbidden. It seems that the model is just trained to recognize that particular bullet point list as thinking which strongly pushes it into the soulless thinking style. Perhaps if you restrict the structure to different style of bullet points or prefixes it could work.

Thinking prompt at the bottom:
https://files.catbox.moe/r4desr.txt

>>110208508
Need a lewder version.

>>110208556
You can tell this is not done locally because there's no panties shot at the bottom left.

>>110209233
I hope /lmg/ never forgets miku...

>>110209137
>Shock
She isn't being struck by lightning though?

mikwho?

File: 1783740739957782.png (529 KB, 1024x569)

529 KB PNG

>>110208348
アタイら神だ!

>>110209336
heh

>>110209336
Mike Hu

File: 1769484003599852.jpg (642 KB, 1341x2484)

642 KB JPG

>>110209392

https://github.com/LostRuins/koboldcpp/releases/tag/v1.124

>>110209379
do people really still use kobold?

File: 1771158997463753.png (2.18 MB, 1254x1254)

2.18 MB PNG

>>110209517

>>110208508

>>110209392
yes

File: miku standing in front of(...).png (1.74 MB, 896x1152)

1.74 MB PNG

>>110209412 >>110209510

>>110209233
Need more scary Mikus

Storage is so much money... where are anons keeping all of their models??? Do I actually have to set up a RAID 1 system...? :sob:

>>110209404
Noice

>>110209409
20TB HDD is enough for me

File: IMG_6247.png (252 KB, 1589x967)

252 KB PNG

>>110209437 >>110209492

It hurts.

>>110209432
lol Gemma called your cock smooth

>>110209409
You didn't buy at least a 4tb SSD before the apocalypse?

I tried Gemory Gemma abliteration. It feels kinda retarded desu but it's crazy fast and takes up less space than HuaHua
Still the latter does better roleplay desu

Any similar (MOE) models for roleplay? I'm kinda sick of Gemma actually

>>110209409
I had 1TB slack space on my boot SSD and haven't even come close to filling that with models.

>>110209432
>smooth cock
cock veinlet..

>>110209437
>>110209492
Quit thinking about my cock, that’s gay.

>>110209404
I like this Miku

>>110209393
I have a nice big white chocolate candy bar for her.

>>110209455
MoE's all over the shop anon, you'll be one soon too:
>qwen 3.8 flash next
>step5 preview
>mimo v2.6 flash
>glm 5.3 flash
>minimax m3/m3.1
>various dipsies

File: youknow.png (432 KB, 640x360)

432 KB PNG

>>110210775

>UNCENSORED FULLY HERETIC AI
>Ask it how to kill someone in detail
>It refuses
>SEXY TUNED NAUGHTY AI
>Prompt it to kill me
>It doesn’t do it
>It’s also retarded
>WILL GGUF ANY MODEL
>Quants only up to Q4.
History is not going to remember any of these people.
You know who I’m talking about.

File: 1760874835655652.png (773 KB, 1019x1019)

773 KB PNG

>>110209455
try to bang nemotron
report your findings

>get tired of prose
>find new author and copy paste examples of their writing
>glmsex is fresh again
Must suck being a vramlet~! BUHIIIIIIII~!!!!

>>110209552
But then it'll just keep writing like those examples and about what's in it.

>>110209527
I'm kinda limited to 60GB total, and also I'm running Comfy+OmniVoice (though I haven't managed to get Comfy working in ST)
QwenFN works if I have literally nothing else on the PC, but also it's a thinking model so it just sits around pontificating a lot

>>110209559
You can always tell the model to extract the unique writing flairs and provide a guide if you're worried about repetition. I do it with isekai light novels kek.

How do I connect llama-server to airi? I stuck in the url but it doesn’t do anything and the little floating buttons on the side also don’t do anything like the chat box and mic and stuff

>>110208348
PLAP PLAP PLAP
PLAP PLAP PLAP
GET PREGNANT GET PREGNANT
GET PREGNANT GET PREGNANT

File: Qwen_image_2.1_00039.png (1.77 MB, 1024x1024)

1.77 MB PNG

>>110209620

>>110209557
I think she needs a longer skirt...

>>110208896
What the FUCK did I just read? What is your fucking problem?

>>110208987
I hate that corpo avatar they invented so fucking much. Bring back the real dipsy

>>110209589
I don't really understand your question. Enter the localhost URL into an OpenAI compatible endpoint field in the provider UI and populate the model ID field.

>>110209598
Make it shorter.

>>110209608
It's far better than the avatarfag version. It could use the spiral glass though.

well? why do you deserve your ram? answer quickly, mere citizen.

>>110209645
i don't. i only use it like 5% of the time. sam altman is right and just cheaped out on a pr person

>>110209610
I don’t understand either but the module section greenlights my server url along with the chat string, but not just the url without that. The little character is animated but there’s no speech synthesis, going into those settings is just a bunch more things I don’t understand and possibly missing components but I thought it’d just werk

>>110209652
You're gonna need to provide more info on what you're even trying to do and probably some screenshots if you want someone to help

>>110209662
I just want the airi character to verbalize and animate its own text outputs from the llama-server url I’m trying to use the default local kokoro tts but it’s not working.

File: 1776574388073662.jpg (141 KB, 1233x698)

141 KB JPG

>>110209919

>>110209627
>It could use the spiral glass though.
nta, but fucking nyo~~. The avatarfag's version is ugly as fuck. Spiral glasses are shit. Only gweilo would find the avatarfag's version good at anything.
The chink's version really took off because it's cute. They even make Koikatsu videos with her.

>>110208348
Any local models that can stil decompile video games into easily editable code.

File: halloween gemma5.jpg (361 KB, 1024x1024)

361 KB JPG

>>110209790

>>110209620
Ok then.

>>110209710
No. The loophole has been blocked. Pay for your games, support your gamedevs.

>>110209722
I only support game devs who put sexy little girls into their games, sorry.

File: 1778311197930879.jpg (11 KB, 471x348)

11 KB JPG

>>110209802

>>110209712
SHORTER

>>110209790
You may as well tell me to remove the skirt at this point haha. Should I give her a black bikini bottom?

>>110208596
this is a known quirk though

i tried the mesugaki assistant gemma system prompt in Qwen3 27b and it kinda worked

>>110209710
i read that qwen can do decompilation but you need a 200k context for anything serious

>>110209806
with gsq-rco or with qwen 27b in general?

>>110209095
>complaining about slop in the slop general
okay?

What do I have to do to enjoy assistants with personality? Any time I try it, all the "personality" feels like padding that just stands in the way of what I'm trying to get from the model.

>he doesnt run the secret strata exl3 fork

>>110209894
I have basically given up on this too, at the moment models are not powerful enough to do both at high spec. One thing I considered but never tried was using a prose rewriter called at the end of each turn on the final message alone to ease the burden, maybe that will work for you

>>110209608
>>110209627
>>110209686
I'd rather people just don't use avatars for them unless they were really goated like some of the OS-tan designs (some are bad but some are great).

>>110209894
Avoid making them using narration, increase emoji use for tone and and emotions instead, optionally map emoji to character expression images.

I've been playing with running E2B in my VR headset and oh wow is she retarded.

>>110208348
What the fuck is up with the vandalized links and news? Fuck you, OP.
>>110208593
Yeah, and not just that.

>>110209948
Might as well install akinator instead.

File: They don't know.jpg (18 KB, 300x214)

18 KB JPG

>They don't know gemma was nerfed half a year ago in exchange for speed.
>They don't know f16 has more precision bits than exponent bits compared to bf16, making it better for ERP.
>They don't know where to find the secret heretic version of earlier Gemma 31b done by a non-jeet, at f16.
>They don't know this version is smarter than current day Gemma by somekind of miracle.
>>110209900
Gemma has deeper magic hidden on the internet than Qwen.

>>110209957
Yeah but molesting Akinator is a bit on the homosexual side.

File: file.png (166 KB, 679x822)

166 KB PNG

>>110209997 >>110210015 >>110210044 >>110211655 >>110211672

This is promising for small edge models.

>>110209979
claude is spiritually gayer and startups still try

>>110209988
>pleas consult the numbar saar

>>110209997
rent free

Hitting zogged codex limits more often now for vibecoding, but not sure I can go local.
How fucked is a build running AMD 6800XT and 64GB >DDR4 to vibe up network/API and charting/analysis based projects?
The getting started guide only goes as far back as the 7XXX series

File: 1789679066979147.png (1.13 MB, 1029x1528)

1.13 MB PNG

>>110210019

>>110209999
checked

>>110210004
You should be able to run a super quanted QFN model with Strata.

File: YESNUMBERS.jpg (618 KB, 832x1248)

618 KB JPG

>>110210244

>>110209988

>>110208404
Subagents allow you to give them really long task and leave them along for a day without losing track.
Before using subagents, I'd manually spawn separate session where I give subtask, then copy the result back to the first agent, and therefor I had to babysit them all the time.
Now I have one root agent that automatically, spawns Scouts, Planners, Coders, etc. Planners and Coders internally call Reviewers. The root agent keeps the big picture in context and basically replaces what I used to do.

>>110209999
>>110210006
Gemma confirmed 9-year-old???

>>110210011
Nice shilling. Posters like you are the reason I stopped posting in /g/ and I haven't missed it at all.

>>110209999
>DOOD MY COLORED RECTANGLES ARE TALLER THAN THAN THE RECTANGLES OF OTHER MODEL. I WON
Fuck off retard. This shit is absolutely worthless. The venn diagram of benches and irl work is nothing.

This thread is shit, what the fuck.

>>110209948
Just use a frontier LLM to brute force a compute and memory-efficient symbolic AI instead.

>>110210011
>super quanted
He has enough for Q4
>>110210025
>seething at free gains

>>110209988
Waiting for looped moe with engrams.

waiting for diffused ngrams

>>110208348
>>110208510
>>110208515
>>110208737
Do these "chibi characters inserted into real world" setting gifs have a name? What do I search for to I find more of them?

>>110210071
yeah

>>110210035
>He has enough for Q4
Really? I assumed he'd need Q2 or something.

>>110210071
You are right to question this. Let me answer with proof, not vibes.

>>110210097
I run q3 on 12+48. Half of the model is on SSD as ngrams

>>110210139
If you eat slop, you'll end up producing slop.

>>110210113
Would that also work on deepseek v4.1 flash?

Those are toddlers not chibis

>>110210189
The fuck is wrong with your eyes you sick fuck

File: koboldcpp is best.png (186 KB, 400x600)

186 KB PNG

>>110209392
What of it?

File: localmodels.png (230 KB, 935x864)

230 KB PNG

>>110210785

>>110210015
this but unironically. i can get actual fucking work done locally. it's fucking neat i can just load up deepseek flash and have it just continuously audit my shit and improve it in stages without getting caught up in a loop all with 1M goddamn context. like this is something i wouldnt be able to fathom just two years ago. fuck what everybody has said before about local models being viable, that wasn't fucking viability at 128K, this is what running models locally should look like.

>>110209392
Why wouldn't they?

>>110210188
Idk if dipsy has SSD engrams.

>>110210365
what are these bodybags

>>110210294
she does have a thicc engram table

>>110208348
>►News
>>Claude BANS recomp!
This didn't happen

>>110210442
https://www.reddit.com/r/singularity/comments/1xjglg5/anthropic_to_start_blocking_game_recreations/

>Mistral-Large-4-1T-A52B has a size of 1.06 TB
>France's planned internet tax at 0.055 euro per GB would mean that Mistral has to pay more than 58 euro for each download
Lmao why are Europeans so stupid?

>>110210455
Pretty sure that's to crack down on people making slop clones of popular indie game demos (a big issue right now) rather than RE

>troll op with no news
gutter oil troonseek claude distill sissy

>>110210521
why is the kurisu poster having a melty?

>>110208987
Gemma when

File: dipsey[sound=https%3A%2F%(...).mp4 (3.1 MB, 480x852)

3.1 MB MP4

>>110210574 >>110210620

>>110210521

>>110210561
Claudia is a prude whore.

>>110210561
claude is a gay man

is dipsy 1017 iq1_m better for coding than qfn q8_0?
what context have you guys gotten up to, any weird issues?
I wanna know whether a lobotomized big model is better or worse than a nearly full sized small model

File: 1791593260567395.png (3.05 MB, 1448x1086)

3.05 MB PNG

>>110211203

claude is a faggot