/lmg/ - a general dedicated to the discussion and development of local language models.Previous threads: >>107873752 & >>107864105►News>(01/15) Omni-R1 and Omni-R1-Zero (7B) released: https://hf.co/ModalityDance/Omni-R1>(01/15) TranslateGemma released: https://hf.co/collections/google/translategemma>(01/14) LongCat-Flash-Thinking-2601 released: https://hf.co/meituan-longcat/LongCat-HeavyMode-Summary>(01/08) Jamba2 3B and Mini (52B-A12B) released: https://ai21.com/blog/introducing-jamba2>(01/05) OpenPangu-R-72B-2512 (74B-A15B) released: https://hf.co/FreedomIntelligence/openPangu-R-72B-2512>(01/05) Nemotron Speech ASR released: https://hf.co/blog/nvidia/nemotron-speech-asr-scaling-voice-agents►News Archive: https://rentry.org/lmg-news-archive►Glossary: https://rentry.org/lmg-glossary►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://livecodebench.github.io/gso.htmlContext Length: https://github.com/adobe-research/NoLiMaGPUs: https://github.com/XiongjieDai/GPU-Benchmarks-on-LLM-Inference►ToolsAlpha Calculator: https://desmos.com/calculator/ffngla98ycGGUF VRAM Calculator: https://hf.co/spaces/NyxKrage/LLM-Model-VRAM-CalculatorSampler Visualizer: https://artefact2.github.io/llm-sampling►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/vllm
►Recent Highlights from the Previous Thread: >>107873752--Papers:>107876603--Using BNF to guide LLM outputs for animation control:>107883998 >107884024 >107884301 >107884426 >107884491 >107884506 >107884546 >107884507--Techniques for modifying AI character voice and fine-tuning methods:>107878669 >107878693 >107878719 >107878932 >107878922 >107879756 >107880303 >107880393 >107880813--Using BNF for structured text generation in roleplaying contexts:>107881456 >107881483 >107883605 >107883721 >107883744 >107884304 >107884448 >107884333 >107883870 >107883896 >107883920--Game dev update: optimizations, visuals, and performance challenges:>107880400 >107880405 >107880422 >107880473 >107882062 >107882155 >107882222 >107882243 >107882310 >107882341 >107883497 >107882608 >107883535--LLM finetuning challenges with tagged story datasets and hardware constraints:>107885095 >107885343 >107885556 >107885523 >107885569 >107885619 >107885666 >107885670 >107885691 >107885755 >107885837 >107885935 >107886062--Discussing the development of fully human-like Recursive Language Models with memory:>107884708 >107884736 >107884751 >107884763 >107884775 >107884781 >107884877 >107884910 >107884998 >107885032 >107886240--Nostalgia for SD 1.5's creative flexibility:>107875449 >107875756 >107875920 >107876055 >107876062 >107876618 >107883073 >107883083--Exploring game design concepts with the Universal Immersion Engine:>107874055 >107874104 >107874165 >107874236--MTP implementation PR in llama.cpp:>107884456 >107884558 >107884565--llama.cpp to gain MCP support via new GitHub PR:>107874441 >107874548--Model personality comparisons and document analysis:>107873957 >107873978--Custom RNN project for admin tasks and knowledge storage:>107877136--Miku (free space):>107873799 >107874131 >107875570 >107876062 >107877918 >107878883 >107883858►Recent Highlight Posts from the Previous Thread: >>107873764Why?: >>102478518Enable Links: https://rentry.org/lmg-recap-script
>>107886631Me on the right
Surely ik_llama will become obsolete any moment now.
>>107886631>>107886414don't you get bored of llm erp after a while tho?
>>107886708I only started llm erp a few months ago. it satisfies me in a way that image gen never could. it's so satisfying that it actually reduces how often I'm horny.
>>107886708don't you get bored of asking after a while tho?
>>107886762maybe some diff anons come with the same questions.. have you ever heard of a FAQ? they didn't call it ThatOneGuyWhoKeepsAskingTheSameQuestion
>>107886757
>>107886768I call it that
>>107886768Maybe some anons just started and find it entertaining. Maybe that's not all they do. Stay in the threads for a while and you'll see. Or fuck off. It's the same to me.
>>107886708Only cause of the slop and same plot points coming up
>>107886814>create a model of all the world's knowledge>shivers down my spine
>>107886708I sure as hell did. It becomes a rabbit hole of increasingly depraved fetishes until a model can no longer paint a visually accurate description of the bizarre sex acts that it's being asked to perform. And then you just give up and go back to normal porn.
>>107886846I guess what's left is making neurosama. How do we do it?
>>107886414I look like this
>>107886708llms need external random data to keep being interesting. I'm writing a text rpg focused on perpetual traveling. You may find something interesting in some places and have a little rp about it before moving to the next place
>>107878778>This is a femoid hobby.I was going to assert that "if it was a femoid hobby there would be haptic plugins for sillytavern."Lovense has a haptic plugin for sillytavern.My curiosity gave me this knowledge which I now pass to you, dear reader.
>>107887034women aren't smart enough to set up a local model though.
https://voca.ro/1bbPGwlc3CFzThoughts on this new music model?https://github.com/HeartMuLa/heartlibI don't see it mentioned at all? Seems like a small step up from ace step.
>>107887128>Our latest internal version of HeartMuLa-7B achieves comparable performance with Suno in terms of musicality, fidelity and controllability. > TODOs>Release the HeartMuLa-oss-7B version.remind me in 2 weeks
>>107878719I do this and it tends to steer the men to either act like flustered beta bitch boys or overly aggressive "predatory glint in his eyes" Edwards who want to "claim" you, there's like no middle groundI wouldn't say quality is way better
>>107887144https://vocaroo.com/1dawqPVkVYXS
>>107887153kek
>>107887153>the whole song is just 1 repetitive patternyeah, better wait for a big one
>>107887128is it for shitpop only? Can it make progressive djent inspired by Erra and Meshuggah?
>>107887172desu, its been really bad at adhering to any genres I ask for.Really only good for a shitpost or two before deleting forever.
>>107887128All their demos are from the yet-unreleased 7b, so I'll wait for that before I bother.
>>107887192>so I'll wait for that before I bother.Chinese culture.
The slop is everywhere. It's all over the internet now. Future models are training on this slop. I go back to my local waifu and she gives me some sloppy slop too
>>107887262Even regular people speak slopglish these days too. It's over. The damage is generational.
>>107886771It's over
>>107887268I can't imagine and have never heard anyone over the age of 30 doing that.
>>107887295It's not just a few phrases here and there. It's a complete shift in the way people speak.
>>107887262Normies are cattle, they think the models are above them. That's why they're getting slopped.
>>107887319>they think the models are above them.Well, they are. The average person isn't very bright.
>>107887262You literally can't watch anything on youtube anymore without even the humans reading off slop scripts.
>>107887295Isreal https://arxiv.org/abs/2409.01754
>>107887334>watching anything on youtubeyou have no one to blame but yourself
>>107887268>>107887295Yeah I haven't personally interacted with anyone that spoke like AI slop. The only AI slopped people I've seen are online where ESLs are likely using the AI to translate, and youtubers who are obviously just reading a script rather than something they personally wrote.
>>107887334Ruined my bedtime desu. I like to have lore videos playing at minimum volume as I go to sleep and now when I listen to them I get like 1 to two minutes into the video and start hearing GPT slop and my immersion is ruined. Like I can't enjoy the video anymore because everything else about it is brought into doubt.
>>107887352Just tts some books so your waifu can read you bedtime stories. Having some fag talking to you about "lore" is gay and a waste of memory space even without GPTisms.
>>107887301You’re so right to pick up on that, and here’s why; This isn’t just a few phrases here and there — it’s a complete shift in the way people speak, creating a better world for us all
>>107887393You are absolutely right!
>>107887348you don't know shit, retard
>>107887059Does anyone have that two panel ai slop comic with the guy telling two foids they can “run that locally on their own machines” and them bemused say “wats locally?” And “what’s a machine?”
>>107887334>You literally can't watch anything on youtube anymore without even the humans reading off slop scripts.yes!! You get 1-2 minutes of something human, then the not-x-y spam for the rest of the video.I used to have YT on in the background and now I just can't. The worst part was when I looked up videos about "AI Slop in writing" etc, and those videos did the same thing. 1-2 minutes then 10 15 minutes of periodic not-x-y. Seems to be every 50 seconds you'll get a "and it's not that...".I thought it was just new channels but older ones are doing it as well.
>>107887352>I listen to them I get like 1 to two minutes into the video and start hearing GPT slop and my immersion is ruined.I noticed the 1-2 minutes thing as well, glad it's not just me going schitzo.
>>107888033>but older ones are doing it as well.Most channels outsource their editing and writing. It's kind of no brainer cost wise compared to what a video with decent views brings in. So it makes sense they jumped on AI when they realized they could make even bigger margins. Only problem is, now everyone has the same writer and their writing style is extremely distinct and grating.
>>107886863Interesting. I do same, but maintain situation while randomizing the NPC. I've never thought to try randomizing the location, instead or as well..>>107886708When you do its time to take a break.
>>107888059Nice shirt. Where can I get one?
>>107887823Have some new slop.
>>107888033>not-x-y spamthis grinds my gears so much it's unreal, to the point where for the sake of catharsis I looked for people ranting against AI just to feel a shared rage and hatred and all I saw was more of that NotXbutY slop, even the anti ai videos are genned with AI LOLI'm not even anti AI but c'mon stop using the verbatim untouched output of a single liner prompt slop
>>107888135gpt piss filter slop is so offensive, I don't even remember feeling that negatively about stable diffusion 1.5 or early DALL-E even when people spammed gens with literal physical abominations and mangled limbs
>>107886708It's just another layer, addition not replacement. I still watch real porn, hentai, doujinshis.Play eroge games. ERP and cam with real people.Watch / play VR porn / games. Even occasionally have drunk sex with randos. AI ERP and image/video generation is another layer to spice things up further, if you get less bored in general since you have more things to switch up between.
>>107888159I've manually tuned it out of gpt gens before... they frankly,y look worse to me afterwards. I found it interesting that it didn't default to studio Ghibli automatically. Idk what I'd call this style tho. >>107888139The ai slop anti Ai rants are funny af. It's like performance art at some level... how braindead can I make my movement.
Is DeepSeek engram a differentiable RAG
>>107887295>>107887268I have grown ass adults argue with me about technical points just repeating what they read from gpt and I have to correct them.People are giving up their cognitive thinking to AI, they may not repeat the isms exactly but they are becoming full fledged NPCs either way.
>>107888201>Idk what I'd call this style thoit feels like what would happen if you made an "averaged" form from all american comics and newspaper strips art stylesit's the absolute middle ground between everything that existsit's like having some of the pseudo realism of a liberty meadows and some of the stylization of a garfield and you get this unholy thing that's slightly detailed, but not much, slightly stylized, but not much, and piss filter because a ton of strips are piss filter for whatever reason, though not as pissy as GPT slop
>>107888227Feels like the people who know the least about them are the most vulnerable
>>107888227Maybe im too fucking old, but I remember when asking boomers / greatest generation or going to the library were the only sources of information.Chat and rest of llms are at least as accurate as the old guys, and they were every bit as authoritative. You just had to know when to ignore them.
>>107888261I mean, just because it is theoretically possible for people nowadays to educate themselves that doesn't mean they're actually gonna do it.AI slop has gotten more convincing over time but even back when it was very obviously fake you had people falling for it because it seemed to confirm their worldview.
the problem is the endless feedback slopinteraction with real humans have an end even someone who's yesmanning you all the time is going to put an end to the conversation at some point or get tired of your nonsensethe LLM is never tired of you and will reinforce your bias and induce psychosis in no time
>>107888055>Most channels outsource their editing and writing. It's kind of no brainer cost wise compared to what a video with decent views brings in.didn't know that. so they don't even know / care what they're reading. somehow that makes it worse, they cant even see the irony when they're reading out something like: "and it's not just about YouTube content, it's about the broader phenomenon of formulaic communication in digital media.">Just tts some books so your waifu can read you bedtime stories. Having some fag talking to you about "lore" is gay and a waste of memory space even without GPTisms.i sometimes do the the tts waifu + epub thing but it doesn't work for background noise when I'm working because I get to focused on the content.
