/lmg/ - a general dedicated to the discussion and development of local language models.Previous threads: >>107686942 & >>107679732►News>(12/26) MiniMax-M2.1 released: https://minimax.io/news/minimax-m21>(12/22) GLM-4.7: Advancing the Coding Capability: https://z.ai/blog/glm-4.7>(12/17) Introducing Meta Segment Anything Model Audio: https://ai.meta.com/samaudio>(12/16) MiMo-V2-Flash 309B-A15B released: https://mimo.xiaomi.com/blog/mimo-v2-flash>(12/16) GLM4V vision encoder support merged: https://github.com/ggml-org/llama.cpp/pull/18042>(12/15) llama.cpp automation for memory allocation: https://github.com/ggml-org/llama.cpp/discussions/18049►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: >>107686942--Implementing character roleplay with system prompts in Open-WebUI to constrain LLM responses:>107697050 >107697202 >107697800--Quantization quality thresholds in Llama.cpp for large language models:>107694813 >107695987 >107696086--Quantized model optimization under RAM/VRAM constraints:>107688512 >107688542 >107688581 >107688771 >107688839 >107688911 >107689227 >107689299--Feasibility of running 4.7 model with 128GB RAM and 32GB VRAM at 3T/s speed:>107694348 >107694375 >107694574 >107694605 >107694687--RWKV.cpp as Microsoft's on-device AI implementation:>107697911 >107698596 >107698971--Open-source model GLM-4.7 achieves top ranking on benchmark index:>107689325 >107689538 >107689545--M2.1 model performance and roleplay evaluation:>107698092 >107698194 >107698171 >107698182 >107698198--Hardware selection dilemmas for local LLM enthusiasts:>107687115 >107687159 >107687197 >107687217 >107687326 >107687392 >107687421 >107687348 >107687388--Quantizing Llama model with bf16 tensors:>107696219--Tennessee AI training restrictions on emotional relationships:>107698160 >107698180--Gaslighting language models to bypass censorship:>107692222 >107692310 >107692314 >107692485 >107693252 >107695957 >107696118 >107696260 >107696518--Browser-specific performance differences in ComfyUI workflows:>107695920--Anticipation and skepticism around Small Creative:>107689009 >107689037 107689080--AI tech for authoritarian parenting in China:>107698384 >107698600 >107698985--AI as interactive fiction game director:>107690487 >107690540 >107690553--Proposing Cockbench update with chat templates for training insights:>107698263--EGPU scalability for local 3T parameter models:>107692736 >107692864 >107692886--Miku (free space):>107688568 >107690307 >107694744 >107688652►Recent Highlight Posts from the Previous Thread: >>107686945Why?: >>102478518Enable Links: https://rentry.org/lmg-recap-script
Gemma
>>107700893Why was this hit?
>>107700924Canceled
>>107700977Previous thread links were wrong. Unacceptable.
is local still light years behind in TTS voice cloning?
the ai is being very kind and understanding. im inspired to embrace my inner snowflake
>>107700909Bald was better
>>107701174this is illegal in TN
>>107701174There is no wrong way to fantasize
>>107701174Yes, even the very AI the ones who make the models try to suffocate agrees with you and not them.
>>107701253There is, it's literally part of "wrongthink".
>>107701268i want to migu migu
>>107701088>is local still light years behind in TTS voice cloning?local TTS is light years behind local voice cloning was perfected pretty early on with RVC models in my opinion. If you can get a TTS from some other source, turning that voice into someone else's is trivial locally at that point
>>107701174What model?
>>107701310>huggingface.co/google/switch-c-2048
>>107701314tank
>>107701280migu is not for migu
>>107701310there aren't a lot of models that allows you to fuck the still warm corpse of a dead eight year old little girlit's nemo 12b or specifically rocinante v1.1
>>107701332Kek, nemo is pretty horny to begin with
Is Gemma 3 the best general purpose model around 30B parameters, or should I use something else?
>>107701332>corpseHow does it work? She literally can't react
>>107701394Roleplay, it can still describe you what's happening.
>>107701386Gemma3 feels so cucked to meI prefer qwen3-vl 30b
>>107700909
>>107701394i like to have another character watching me do it too, it's even better if it's a woman describing her horror over what im doing, along with the dirty detailslike when the dead little girl's bladder releasesim such a sick fucker...
>>107701450Maybe Tennessee is right after all...
>>107701332>rocinanteplacebo>nemoonly reason it was good is complete lack of censorship. makes me wonder how much better glmchan would be if she had zero censorship.
>>107701534Sucks that Chinese models can never have zero censorship since they just distill it from western models. Hopefully one day they can move past that need.
>>107701332Are you using straight nemo or some kind of finetune of it? It always seemed to lose coherency for me, maybe I need to give it another shot.
>>107701433Fixed your glazed garbage ^.^
>>107701332>it's nemo 12b or specifically rocinante v1.1How does it compare to normal nemo? Like what benefits?
Is Medusa Halo going to save local models?
>>107701631UOH?!?!?HOW DID YOU REMOVE THE POISON?!?!
>>107701715>OH NO MY ART>I better make it look like shit, that will show the AI people!!!
>>107701696>Like what benefits?It got astroturfed when nemo was new. People downloaded it instead or regular instruct and thought it is the magical finetune and not just the instruct model being good.
>>107701696You can use ChatML with it if you're really anal about using the format. That's about it.
I actually went back to 4.6 for sex.
