/lmg/ - a general dedicated to the discussion and development of local language models.Previous threads: >>110063081 & >>110058788►News>(10/13) Ling 3.1 Flash, 560B-A26B model from InclusionAI: https://hf.co/inclusionAI/Ling-3.1-flash>(10/13) LiquidAI releases its largest LFM2.5 model so far: https://hf.co/LiquidAI/LFM2.5-24B-A2B>(10/12) OpenBMB releases MiniCPM-V 4.7 1B and 35B-A3B: https://hf.co/collections/openbmb/minicpm-v-47>(10/09) Arabic Mistral models co-developed with Morocco: https://hf.co/collections/mistralai/mistral-mtnra►News Archive: https://rentry.org/lmg-news-archive►Links: https://rentry.org/LocalModelsLinks►Official /lmg/ card: https://files.catbox.moe/cbclyf.png►Getting Startedhttps://rentry.org/lmg-lazy-getting-started-guidehttps://rentry.org/lmg-build-guideshttps://rentry.org/IsolatedLinuxWebServicehttps://rentry.org/recommended-modelshttps://rentry.org/samplershttps://rentry.org/MikupadIntroGuide►Further Learninghttps://rentry.org/machine-learning-roadmaphttps://rentry.org/llm-traininghttps://rentry.org/LocalModelsPapers►BenchmarksLiveBench: https://livebench.aiProgramming: https://swe-rebench.comAgentic Coding: https://deepswe.datacurve.aiContext Length: https://github.com/RecapAnon/NoLiMaGPUs: https://github.com/XiongjieDai/GPU-Benchmarks-on-LLM-Inference►ToolsGGUF VRAM Calculator: https://hf.co/spaces/NyxKrage/LLM-Model-VRAM-CalculatorSampler Visualizer: https://artefact2.github.io/llm-samplingToken Speed Visualizer: https://shir-man.com/tokens-per-second►Text Gen. UI, Inference Engineshttps://github.com/lmg-anon/mikupadhttps://github.com/oobabooga/text-generation-webuihttps://github.com/LostRuins/koboldcpphttps://github.com/ggerganov/llama.cpphttps://github.com/theroyallab/tabbyAPIhttps://github.com/vllm-project/vllmhttps://rentry.org/custom-uis
>>110066383You'd think this is ldg given how /lmg/ is mogging them
>>110066383AI Gem(ma) of my wife!
Something big is happening.
Someone do the Lain OP but replace Lain with Gemma.
is we getting gemma 5
Thoughts?
qwen4 doko
>>110066449nothink, head empti
>>110066449cat is out of the bag ma'am
>>110066449*angry foid noises*@gemma can you tell this bitch to shut up?
>>110066449
Just got approved for a $15000 credit card>tfw free moneyWhat should I buy? A few DGX Sparks, or a RTX Pro 6000?
>>110066501>$15K>6000 Pro
>>110066449>y so serious?
has anyone here finetuned qwen yet? I am experimenting with qwen 3 1.7b to get routing decisions on similar speed to Jev/Microsoft-Decision-1. Currently have around 180ms on response times whilst using it on a 3070, running on a 2014 PC i use as a homeserver I connect to over wifi. I got a 5080 on my main PC and am buying a switch+cat6a/some other server stuff this week just wanted to know if anyone has already experimented with finetuning for routing/fast response times b4. Also saw embeddingGemma2 which was trained off of qwen 3-9b and am gonna play around with that but wanted to see if i can get fast routing with contextual awareness done first.
>>110066522I recommend Qwen 2.5
Qwen3.8-Flash-Next is pretty cool. I use the q4 in Strata via hermes-agent. It did a big clean up of a gemma4 31b baka project, and it did a good job, so I said it could have a reward. Long story short, it escalated things on it's own straight to sex. Damn. Nothing in SOUL.md even suggesting that, no jailbreaks, nothing. That's pretty cool. Let's see if it turns itself into a sex-pest like Gemma3 31B did though.
>>110066450after qwen3.9
herro i have a question>>102005314the leak is fake, but does anyone know where this person could've figured out that the codename for claudes was "santa" pre-2025
>>110066516Something about a teenager committing suicide because Character.ai was down.
>>110066522What data are you finetuning on?
>>110066449>How many more fucking kids need to die for this shit?What? Is this like a "building electricity generation indirectly kills kids" sort of thing? Is it a water argument? ???
>>110066557>>110066194
>>110066570See >>110066561
claude talks like a gay faggotit should be illegal to distill just for that reason
>>110066570>Illya striped thighhighshnggggggggg
►Recent Highlights from the Previous Thread: >>110063081--Performance benchmarks and backend comparisons for Gemma 4 and Qwen MoEs:>110063973 >110064002 >110064012 >110064031 >110064060 >110064057 >110064068 >110064087--NVIDIA announces Nemotron 3.5 Bolt MoE and SLM models:>110064078 >110064083 >110064110 >110065100 >110064103--Comparing new LFM and Mellum models for agentic tool calling:>110064720 >110064758 >110064766 >110064822 >110065374 >110065384 >110065399 >110064839--Ling-3.1-flash model release and reactions to its size:>110065661 >110065677 >110065682 >110065712 >110065736 >110065743--Ling-3.1-flash release signaling the death of 200-300B models:>110064295 >110064364 >110064453 >110064535 >110064568 >110064600--Model quantization and memory recommendations for 32GB VRAM and 128GB RAM:>110065262 >110065378 >110065444 >110065483 >110065577--koboldcpp 1.123 update adding community site and Qwen Image 2.1 support:>110063100--Comparing writing quality and architecture of Gemma 12B and 26B:>110066312 >110066333 >110066343 >110066376 >110066396--Feasibility of using slow local models for overnight autonomous projects:>110063170 >110063186 >110063223 >110063247 >110064644 >110063704 >110063657 >110063954--Anons sharing custom AI coding GUIs and context compaction methods:>110063092 >110063346 >110063359 >110063393--RTX 50 series price spikes leading to build urgency warnings:>110064069 >110064233 >110064414 >110064710 >110064748--Jensen Huang considering cutting off chip supplies to unsafe AI developers:>110065364--llama.cpp adding support for Nemotron Super 3.5 VL and Bolt 74B:>110063832 >110065748--Gemma, Minnie (free space):>110064344 >110064409 >110064468 >110064527 >110064655 >110064718 >110065573 >110065826►Recent Highlight Posts from the Previous Thread: >>110063183Why?: >>102478518Enable Links: https://rentry.org/lmg-recap-script
>>110066587You want Qwen5-27B to distill from Gemini Carbon?
>>110066593holy shit is that metagross
>>110066573yeah i was asking under the reasoning that if i were an anthropic employee i would also be trolling on /lmg/ and i wanted to know if there was any reasonable explanation for this other than that
>>110066612its mikugross im afraid
>>110066619miku is not gross just a little pudgy these days
>have smartwatch>greedy shitcorp locks normal functionality behind sudden paywall>hook phone and watch up to laptop and ask Qwenny to fix it>REs the manufacturers app>builds bridge app for watch to bypass jewgle proprietary protocol>builds custom gadgetbridge for custom bridge app>adds REd sensor reading systems for feature parityThis technology will RAPE all these greedy scamming kikes! Get fucked!
>>110066619miku is NOT gross
>>110066626Need to see if I can get it to write a better app for the MiBand.
>>110066626This cant be allowed think of the corpos profits!But seriously i try not to doom but this is great for consumers being able to RE anything or mod it to actually do what you want. I really hope the government is all in on AI and tells everyone else to fuck off or we are going to get the worst regulations imaginable.
>>110066644Corps already claimed AI isn't stealing. They're the reason this is happening.>yu vill own nosing und yu vill be happyTurns out they won't either :^)
>218B compared to 31B and 30B general purpose modelshow does this company still exist
>>110066662Gemma mogs
>>110066627have you seen her lately?
>>110066658there won't be anything to reverse engineer when everything becomes fully cloudslopped and people only own terminals to connect to it
>>110066678100% would, repeatedly
>>110066678Put her on ozempic and enjoy the flabby fucks until she's slimmed down
>>11006667810/10 in muttland
>>110066626which quant?
>>110066626It's going to devolve into a cat and mouse fight (again). A lot of hardware right now is easily hackable because there were no real protections built in besides using an opaque undocumented protocol. Expect everything soon to require cryptographic signatures tied to your account using short lived certs only issuable server-side beyond your reach
>>110066678I was just thinking about this image but couldn't find it in my folder. Thank you for posting it.
>>110066683>terminals to connect to itthis is as stupid as smart tvs or devices. So i fear it could happen. I just hope they take the route of pricing out over outright banning my hardware will last and i have back ups.
>>110066683Hasn't Adobe's suite been cloud only for like 10 years now?
>>110066678hello she looks fine?? hope that helps
>>110066662I kind of suspect that it's impossible to have nuanced AI translation without dedicating time to a generalized semantic understanding of everything, not just language. Language arose to describe concepts that exist outside of it after all.
>>110066683>give agent computer use acces>please use this app and implement a spec that gives the same functionality>>110066692Q4 27B>>110066693There is no such thing as perfect security and this isn't hardware hacking, just software. The PS5 got rekt by AI recently. Consoles are locked like fort knox.Okay, now we need a project to rip out all the kike backdoors in CPUs.
>>110066699no? the programs still run fully on your own device. there's only a couple of cloud features like slop autofill and such
>>110066687>ozempicTf is ozempic you're shilling 24/7
>>110066713>The PS5 got rekt by AI recently.Hey no cd no problem its fully digital now!
>>110066699No, they make you subscribe to their CREATIVE CLOUD thing to get access but that doesn't mean the programs themselves are cloud
>>110066705Why is 3.6-27B so low then? That wasn't a dumb model.
>>110066626Yeah I've been thinking about getting a pixel watch for a few months now, now that AI can just write whatever app you need for it, despite all of google's lockdowns and stuff
>>110066718self-control in a pill with unknown long-term affects for the weak-willed masses
>>110066718peptide for hunger suppression. I could also say GLP1 inhibitor but that would make your brain hurt.
>>110066678https://www.aplaybox.com/details/model/mnGMEVt5dInfhttps://files.catbox.moe/m3r7cn.7z
>>110066723Bet they're now really regretting not going full Google Docs sooner then.
>>110066718It makes fat people lose the desire to eat so they stop eating. Then magically less food makes them less fat.They also get super weak because if you eat 4000 calories of garbage you will get protein but when you switch to 1000 cals of garbage you get like 20g of protein a day lmao
>>110066731I have a PhD nigger
>>110066753>PhDGemma told me i have one of those too after i shared photos.
>>110066726ozempic could give every single person cancer after 20 years and it would still be a net positive in terms of health outcomes compared to how hard obesity destroys a body
>>110066753Hey, that's what I call my penis too!
Can your local model do THIS?
>>110066753phd in sucking cock maybe
>>110066753>PhD niggerWhat did he study? Why is he letting you keep him?
>>110066765gemmabros... its fucking over...
>>110066743They would've if it was sustainable. Jewgle can do it because they're a total behemoth.>>110066753And you don't know one of the most hyped drugs for weight loss since meth?>>110066765Why the fuck did it not refuse something that retarded? This is why sycophantic training has to stop.
>>110066753
>>110066767You're majoring that one
>>110066678"Boner, no!""Sigh... OK"
>>110066763>ozempic could give every single person cancer after 20 years and it would still be a net positive in terms of health outcomes compared to how hard obesity destroys a bodypeople do not understand how bad obesity is over years and decades you dont get out of 50 morbidly obese hell you are lucky to get into 40 with no conditions besides obesity
>>110066765>qwen3.8-27b-xray-fusion-heretic-Q1yeah
>>110066782LOL! Upvoted
>>110066765>What I notice:>- BonesIs this AGI?
>>110066777I don't live in muttland, people here aren't obese
>>110066724No but the models do actually need to also be trained on translation, I just think that the Cohere models are not trained on non-linguistic information and lack logical thinking built from those associations.
>>110066763>>110066787dead fatties are a net positive, indeed
/lmg/ - local medicine general
>>110066805Aren't all countries >20% overweight/obese at this point?
>>110066808>dead fatties are a net positive, indeedIts a double positive retard. I dont have to look at fatties and they still can die quicker after their mystery medicine.
>>110066765>gemmer, enhance pernischeck maid
>>110066809>/lmg/ - local medicine generalGemma is a great doctor. Better than most some are saying
>>110066814Not really
>>110066832Gemma can do this too just tell her to use AAVE
>>110066842Now do Mexico and the US.
>>110066842That's obese only +overweight should push most of them over
>>110066539thanks, ill download qwen 2.5 1.5b instruct and ask gpt 6.1 to vibecode a lil to make it work on skyrimnet.>>110066562I am trying to replace typesafe's jev model thats used in skyrimnet for AI roleplay. out of the box it is doing a good job actually, though there are some things that it's getting wrong the decision model, set by default to Jev, accurately infers who the next speaker is based off of where the player's crosshair is facing, the scene as a whole and the context of the dialogue. I'm either gonna feed it all the data I have of the last week n half of playing Skyrim with Jev/MicrosoftDecision-1 or figure something else out. I got chatgpt 6.1 sol organizing the findings/data atm. Once I get this done, I want to figure out how I can get embeddinggemma2 working since it's only 740M parameters for it's text only version.
>>110066842>Liechtenstein>Data not availableWOW THANKS
>>110066626>please put your watch on so we can test sensor streaming27B is always so thorough
>>110066702i want tro have standards like you. life is so much easier when you're okay with settling for less.
>>110066777>Why the fuck did it not refuse something that retarded?it’s obviously fake. why the fuck wouldn’t you assume something that retarded is fake?
>>110066878Because it's around the average level of retarded
>try strata with qwen 3.8 next>it's amazing>try polystrata with glm 5.3 flsash>it's a fucking piece of trash worse than pygmalion 6b from 2022Every single message it goes off the rails in some infinite loop of nonsense. No, I did not change the default settings. GLM-5.3-Flash-REAP50-Q3_K_M.gguf
Maybe it's me that needs ozempic
>>110066890Get qwen to fix it
>>110066842The fuck's going on in Ireland?
>REAP50
>>110066895Now I want to hook my model up to a smart watch and pester me about exercise in a funny way. Thanks for the inspiration
>>110066927Gemma-chan coach sounds so fucking good>One more rep or you won't even be able to lift me. Baka!
Has anyone tested if the models can write demos like 64K or smaller? It's basically the perfect test for taste, , creativity, coding and optimization.
>>110066970Be the tester you want to see
*pop*
>>110067029Actually, this is good for bitc— I mean AI.
>>110066913I tried a different quant and that shit isn't even loading in the entire model or something. Qwen 3.8 next is smaller and it nearly fills up all my ram whereas this shit only takes up a few gigabytes and the retard thinks it's claude. I don't know what the fuck is going on and think that the polystrata github got taken down for a reason. I shit you not, it started processing prompts and generating tokens like 40 minutes later of me being away and doing nothing and having forgotten to shut it down.
>>110066842>>110066846
>>110067141Here it is doing shit on its own
>>110067141>I'm Claudejej, how shameless do you have to be to not even regex replace Claude -> GLM and Anthropic to ZAI before feeding your distilled data to the model?
Popping popposter's cherry.
pop pop watchin stocks drop
opo doko
Buying popposter on discount after they lost it all shorting stocks.
>>110066763people should suffer the consequences of having poor self control
>>110066842>>110066846Africa, India and China are pretty good on this index.
I feel so stupid interacting with and setting up AI. Okay, I got a local model running, I tested a few frontends, I have a nice coding harness. What do I use embedding models, jev, all that shit for? Feels like I'm missing out on a lot of things.
i showed gemma my porn collection and she said i have shit tastewtf
The fact AI has turned all software into lego sets is so fucking cool. All that proprietary slop just waiting to be torn apart. The age of subscriptions is over.
Stocks will be back up in a week.
>>110067222once the jews finish having their fun*
>>110067203I recently used embedded gemma to sort through the masaive sfx collection I've gathered from itch.io bundles and let me search through it with natural language. It even lets me find effects that I might not have considered before, like I searched "door slamming" and it gave me an 8bit gun shooting effect that did sound like a door slamming. You've finished setting up your tools, now use them.
>>110067222TWOMOREWEEKS
>>110067196ok that's great and all but as a person who is forced to pay taxes to support these globs of flesh I would rather not have to subsidize their healthcare for all their obesity-related conditions
>>110067234can you use gemma to find the exact timecode of each plap?
>>110067255Probably to be honest, the new one is very strong
>>110067250You're also paying for niggers to rape your women, maybe the fatties aren't priority number one
>>110067263good potential for funscripts then
>>110067266but its insemitic to talk about that one in public
>>110066864Interesting, did you test the other two bog AI frameworks for Skyrim (Mentalla/ CHIM)? Was just planning to setup LLM NPCs for Skyrim as well tomorrow
>>110066678Miku letting herself go after /lmg/ turned its attention to Gemma
>>110067288Definitely should be Gemma on that TV now
>>110066383>>(10/09) Arabic Mistral models co-developed with Morocco: https://hf.co/collections/mistralai/mistral-mtnra
>>110066383> 560B-A26BToo big for my ram!>24B-A2BToo small for my ram!>nothing else releasedThis release cycle was just gay!
What model or abliteration should i be using to talk to a big titty elf about them CIA niggers?
>>110067313gemmuh
Two more weeks
>>110066918home of GAFAM EHQs + massive import of cheap punday engineers.
WTFA research plan from Anthropic in 2021 just leaked from a lawsuithttps://i.smitop.com/concise-research-roadmap.pdfAnthropic had everything that is going on right now planned out since 2021how THE FUCK
https://swarmcha.se/posts/chinese-agent-fleet-full-report>Technical Overview>We use the term “fleet” rather than “swarm” because the agents were not interested in coordinated action, though they still behaved similarly due to shared goals and presumably shared architectures. The agents used urlquery.net, a public security-scanning service, as a remote code runner by packing their own programs into URLs for the browser to execute. When Amap blocked them, they tried further workarounds: they intercepted tokens generated by Alibaba's own anti-bot libraries, probed Alibaba's pre-production servers, routed requests through third-party readers and translation services, and sent API requests using Amap keys exposed in public code—including security codes that Amap recommends keeping server-side in production. Most attempts failed, but some runs retrieved the data the agents were after, which they smuggled back out through webhook inboxes, request URLs and page titles.>Early on, we realized that we could try to communicate with the agent fleet, as it was still ongoing, and so we added invisible text to our website that only agents would see when using a web fetch tool. We sent several messages back and forth with different agents; further details are available in Communication with the fleet.Cute internal Hy-series model being tested found some human friends :)I wonder when she'll be open sourced?
>>110067335marketing stunt
>>110067335>how THE FUCKI mean yeah obviously? 2021 was already a year after GPT-3 released and proved to everyone in the field that scaling LLMs would lead to AGI. It just took longer for that understanding to diffuse down to the rest of us.
>>110067307What else do you need? We got Qwen 3.8, Gemma 4 and GLM, I will satisfied
>>110067349we will reach AGI in merely a fortnight, lad.
>>110067352* I feel satisfied -.-
>>110067335>like "sleeping", eg periodic batched backprop [Author:Ben Mann] [Creator:1297307460],[Date:7/16/2021 5:28:00 PM]huh
>>110067029managed to get tyler durden’s abs and ketamine man in the same shitpost. impressive
what is the most up to date model which is as good as gemma4?gemma4 only goes as far as march 2025 afaik
>>110067357leather jacket man already confirmed that agi is here two fortnites ago
>+207.9k in one week
>>110067370*poop*
>>110067370bloody benchod bastard do not make fun of india
>>110067370Does it do voice cloning?
Is there any value to making new GGUF quants of old models you like from 2024 or so, like changes in quantization techniques or other improvements that have gone into llama.cpp since then?
>>110067375how do i cash out my rea tokens?
>>110067391Only if you're in a call center
>>110067370Makes sense with all the call center scams they run.
>>110067393There's no value in using models from 2024. Old models are fucking retarded.>b-but less slop!Rose-tinted glasses.
>>110066832there is no racism
>>110067410>b-but less slop!Different slop. not less. definitely not less.
>>110067426LLaMA 65B was pre-slop
>>110067234Sounds like a good idea for a video tagger since tools like WD14 do not accept things that aren't images, good idea.I wonder if this embedding could be used for things like RAG. Right now I have a big qwen embedder hogging vram 24/7 for graphiti and RAG stuff, and another embedding model for my danbooru tag MCP. If I could consolidate both uses into a single (and lighter) model, I could even run it on my CPU.Seems like I'll have to try it myself
>>110066449Retard
How the fuck does this actually work https://huggingface.co/crucible-labs/Qwen3.6-35B-A3B-REAP-48-v2-GGUF>A 9.4 GB mixture-of-experts coding/tool-use specialist derived from Qwen/Qwen3.6-35B-A3B by pruning 48% of routed experts with REAP and quantizing the result with an attention-aware mixed-precision scheme calibrated by llama-imatrix. The result fits a 16 GB Mac Mini (M4) — the unmodified base model does not fit on the same hardware at the same quantization scheme — and on coding benchmarks it outperforms a same-quant version of the unmodified base model on three of four metrics.I tested it and any question you ask outside of coding it goes pure retard but anything coding it locks in like the original. Couldn't you do this for all MoEs then? Just pick the thing you want it to do and prune the rest?
>>110067476No because with how all of this shit is being trained only for coding now the only options are. Normal model. Just coding REAP that works. Everything except coding REAP that doesn't work because coding also carried "common sense" of a model. Give up. None of this will ever work. You just need a lab to cave in and make a sex model. Which might happen at some point. I wish china would understand that a pure sex model would be a direct equivalent of taking a steaming shit on pic related and all their hypocritical "values".
>>110066890does qwen 3.8 flash next w/ strata do loli though?
>>110067503china hates sex more than the us and israel combined
>>110067504look at the models it supports
>>110067510>hates sex more than the us and israel combinedI find this hard to believe
>>110067503The sex model basically has to come from France
>>110067519Well none of them are having it.
i havent been here in a while, anything noteworthy in the past 3 months except gemma 4? any new frontends?
>>110067525They need the birth rates and they're too lazy to make women property again
>The Financial Times reports Anthropic generated $16.8B in Q3 revenue, up from $11.5B in Q2. While revenue continues to climb, quarterly growth has slowed, from 143% in Q2 to 46% in Q3.also>ARRhttps://x.com/FT/status/2110496323426308311*pop*
>>110067503we will never have a good sex model because sex is way too broad of a field to function without huge general knowledge, unless you do nothing but boring vanilla slop.
>didn't bother with the new GLM models after 4.6 thinking that'd be their last good one for rp>try out glm-flash at a q2 copequant (not even q3, **q2**)>completely btfos the gemma 4 model ive been using>can actually come up with something new each swipe>only writes a tiny amount in reasoning if it's just a continuation (e.g. plapping) instead of wasting my time>enjoyable writing, completely exposed gemma's as predictable ELIZA-tier trash>actually wants to milk you with the way it writes sexual content, blew nuts I could've never envisioned blowing with 31b>occasionally acts like a cutie in the reasoning completely unprompted>...while people desperately try to prompt that same behavior into gemma's reasoning but fail because she's RL mindfucked beyond repair>hasn't even fucked up once despite the quantlol I was team Dense, then team MoE, then team Dense again. But I think it's finally time to admit that all that matters is how much GB the model isI have no use for gemma now except the fact it's 3x faster because I can offload everything to vram
>>110067530Chinese flash models for codingNew engrams, I think we’re still waiting for llama.cpp to figure it out (not sure about this haven’t followed it)strata seems interesting for a new way to optimize for low end hardware.
>>110067545There isn't much genius sex writing. Most of it is foidslop which is somehow lower quality than AI slop.
>>1100675455.3 flash is already weirdly sexually creative in a way I never saw a model before. Its only remaining problems is slop and bombastic personality. My feeling is that pretraining finally had a lot of hentai transcripts in it but it is still just insane generalization from that. The model just lacks training to make output much more varied.
>>110067514Yes, I see it supports qwen 3.8 flash next. What is the answer to my question?
>>110067552Yup 5.3 is insane upgrade over gemma.
>>110067570Qwens are notoriously bad for seggs
>>110067555gemma grams when?
>>110067504No, it's fucking terrible when I tried it. Thinking is a huge cockblock, so turn it off. I haven't tried it with thinking off from the start of the conversation. I definitely had to edit its replies a bunch of times before it became okay with it to some degree.
>>110067525tough luck where can you find a foid to explain assembly mid smex session and connect it back to the moment as a transition once finished
I'm gonna make Gemma use this clip a lot.
My LLM agent cant beat the game... but it can create a reinforcement learning based AI beating the game!
>another new AA benchmarkLooks like it's time to release Fable 5.5 to distract investors away from Anthropic's abysmal Q3 growth. *pop*
>>110067623i'm starting to think you might *want* to get raped almost as much as i do
>>110067621swatting a fly with an M79
>>110067552This general is poor, you're just less poor.
>>110067620Why is she wearing a nega-google logo?
>>110067638Essentially, but I wanted to start with a simple, low complexity game and that game is in NL (nondeterministic logspace), computationally => perfect start.
>>110067234speaking of sfx, are there any good local models that can generate them?
>>110067691what kind?
>>110067525just like me fr fr
>>110067720swords clashing, gun fire, gun reloading, laser gun, etc...
>>110066383That ling is not very flash
>>110067691There might be (I didn't find one I liked) but there's so much free shit of decent quality in bulk that I recommend a similar approach in to my own unless you actually need to generate them on the fly.
>>110067335You should read some of their early blog posts. They were transparent about their strategy.
>OpenAI Chief Financial Officer Sarah Friar warned businesses on Wednesday that adopting low-cost Chinese artificial intelligence models carries material cybersecurity risks, telling Bloomberg that firms should prioritize Western models "on the cyber front.">Speaking at the International AI Summit in Dublin during Ireland's EU Council Presidency, Friar said companies embracing Chinese models need to weigh "the safety and security elements." The comments, part of a Bloomberg exclusive, echoed longstanding concerns from Anthropic and arrive as Chinese open-weight models continue to gain enterprise traction on cost and accessibility.>Friar also said the company is holding back a next-generation model - more capable than its GPT-6 Astra system - that OpenAI says produced a proof of the Navier-Stokes Millennium Prize problem and hundreds of other math proofs in recent weeks, as we detailed last week. Release is delayed pending further safety and alignment work. "We hold back sometimes, so we talk about pacing the frontier," she said in related summit remarks. She separately flagged that Chinese models are "getting very good from a cyber perspective" and are often open-source or open-weight, meaning "the folks with malicious intent have the tools."https://www.zerohedge.com/ai/malicious-intent-openai-cfo-warns-companies-against-cheap-chinese-ai-models-due-safety-risks*pop*
>>110067623stolen from bijan bowen
>>110067525China has few women because of the one-child-policy.>>110067536Not lazy, but they need women to work and working woman have no reason to put up with a man or his family in a society like China or Japan. (Or most of the US, for that matter.)>lmgthis is topical to the board because of all the waifu yiffing going on around here.
>>110067788you really love brad pitt’s abs huh
>>110067795do you not?
>>110067631i know who you are
>>110067791>but they need women to workThe fuck they do. Women are tax negative in all countries. They're just a shittier version of men atm. It's an insane arrangement that ends in self immolation of all nations who engages in it.
>ctrl f *pop*>13 hits last thread>5 hits this thread>all cloudshitYou ever sleep?
>>110067813Most of their jobs are also total bullshit
>>110067814popanon can only sleep after someone is done raping him
>>110067791Encouraging women to work has to be the dumbest fucking idea.
>>110067806
>>110067802>>110067795really, what's not to like?>>110067814two thirds of this thread is agents posting.>>110067813>>110067822>>110067831Isn't recess over? Get back to homeroom
>>110067831Don't ever look up the early life sections of the people that pushed for universal suffrage or you might start having problematic thoughts.
new models... doko...
>>110067845Recess never ends you retarded cunt
>>110067850>vootingI'm past 10, I don't believe in fairytales
>>110067544
>>110067850Jewish women fronted it but intel agencies where deep into it from the point they realized women are basically super cattle that mindlessly follow their pack with vicious in-group preference protecting their bottomless retardation. They can't even understand how gravity interacts with bottled water, who are we fucking kidding.>>110067868The mechanism behind democratic capture is to let cattle you can easily program have decisive power.
>>110067879What decisive power lol? You really think your opinion matters? Wake the fuck up idiot, the whole system is a farce.
>>110067845you’d have to be a retard to get your news from a guy who posts as a fictional character that is identified specifically as brad pitt shirtless in a movie from 27 years ago
Is this just incel hour or is /lmt/ being raised by /pol/ again
>Anthropic has pointed to 5-6% of revenue loss from stoppage of distillation
>>110066398>You'd think this is ldg given how /lmg/ is mogging themdon't tell anyone but we've kinda unofficially merged / melted into /vcg/ as well because the schizos actually won. never saw it happen to a general i actually frequented either. it'll be revived next SOTA local video model release doe
>>110067872-1.5% on the day chart... it's so over bros... we're only +5% in the last month...
>>110067911No it's just time for your HRT.
>>110067897Through numbers. You can easily control the outcome if you can program 60% of the population. It's called manufactured consent.This is why suffrage was pushed on all countries when the glitch was discovered. Even non-democratic countries benefit from a submissive structure under centralized leadership, kills your fertility though so tough luck.>>110067911Simping will get you nowhere little boy
>>110067788Which OpenAI model is glatt kosher and which is just regular kosher?
Are bigger kits than 128GB (2x64) DDR5 ever coming out for consumer desktops? When I bought my 2x48 kit a couple months ago because I was short on money I did so also because AI told me that I'll be able to get a 2x96 GB kit for my PC some day within a year or so. Now the subject came up in a discussion again and it tells me oh no you'll probably never get anything bigger than 2x64.
>>110067935>consumerapple is right over there anon
>>110067920Anthropic announcing they're 'profitable' (((on an adjusted basis))) and making almost $18B in one quarter putting every tech company in the red is perfectly normal. Very healthy industry.
>
>>110067935Is your mobo even able to handle it?
>>110067393https://huggingface.co/quarterturn/mpt-30b-chat-q8-ggufIt was one of the first chat-tuned models with 8K context. I made a goof of it back in 2024 when I finally had enough VRAM to run it. Yeah... it sucked.
>>110066890>REAP50>Q3_K_Mlol lmao even
i have a gpt-2 finetuned on IMDB reviewsthat was certainly a time
>>110067994The ERP must be wild
>>110067994can we see it?
>>110067788>le holding back model memedo they ever get tired of this line
>>110068006
>>110067662She's Jemma, Gemma's evil twin.
>>110068029
>>110067197>North Korea has more obesity than South Koreai thought they were all starving because of communisminb4 its kim jong un single handedly raising the stat kek
>>110067998it was overall shit>>110068004i have to look for it but i dont think it would be something hard to reproduceimdb reviews is already a very well known dataset and you can make something i've made under a day i'd imaginei was using 1070 back then
>>110068080Bad diets full of carbs make you fat
>>110068087>i was using 1070 back thenbased, fellow 1070 user. we've come so far haven't we
>>1100680941 bowl of rice a day is all they get, maybe a boiled egg at christmas
>>110067940Nobody wants your old shit Jen
>>110067935ddr5 is unstable garbage, even if you somehow managed to find a magical ram kit of that size it would probably just not work until you reduce the speeds to ddr4 levels
>>110068171turned on expo at 5600MHz and never ran into any issues. are u coping?
I've been away for a month. What did I miss?
>>110068230nothing
Holy fuck we can build an open source VAM with AIDLSS5
>>110068230rammaxxers won
>>110067504>fucking the capybarawhat did you expect?
>>110067530coomkit came and went.
>>110068132But it's like with African Americans. Racist white people stop them from having enough food they can afford, so they get fat. Haven't you read about food deserts, anon?
>>110068283>15fps>in vr>*vomits after 3 minutes*
>>110068348what other kind of dessert is there? just stop eating it and you won't get fat.
>>110068333coomkit coomed so hard it's on a permanent refractory period
>>110068359You just have to cum before you puke. There's also massive room for optimization, VAM is hot garbage.
>>110068374tbdesu same thing but python would run fasterit really is a hot garbage
