Welcome to Chatbot AI General #102, the thread for discussing and improving AI pony chatbots.▶ MLP Botshttps://mlpchag.neocities.orgSpreadsheet (CAI bots + Old repository):https://docs.google.com/spreadsheets/d/1J7BeqJVDS51cXF8Pgm2YZaFq-Z6ykSJTCAI bots converted to Tavern: https://files.catbox.moe/ckurq1.zipExpression packs: https://rentry.org/ChagExpressions!!!GALLERY!!!: https://drive.google.com/drive/u/2/folders/1Ao-h5HFGMPllSrzSBKM_BvGSiU9f0c2U▶ How do I start?1) Select a Frontend2) Select an AI model3) Select Jailbreak4) Select bots5) Lovemaking with AI mares!Starting in this hobby can be confusing and difficult. If it’s your first time and you’re lost,▶ ASK THE THREAD! ◀Novice-to-advanced guide: https://rentry.org/onrms▶ SillyTavern (preferred frontend)https://github.com/SillyTavern/SillyTavernOn Android: https://rentry.org/STAI-TermuxApp that voices pony responses in ST: https://drive.google.com/drive/folders/16Ss26VBmgzcSuTGzhaHqRuyVRceTf-YB▶ More frontends:Risu: https://risuai.xyzAgnai: https://agnai.chat▶ Localshttps://rentry.org/lunarmodelexperiments>>>/g/lmg/Mistral Nemo base model fine-tuned on fimfics: https://huggingface.co/Ada321/Nemo_Pony_2/tree/main▶ JailbreaksMLP JB: https://rentry.org/znon7vxeMore JB and guides: https://rentry.org/jb-listingHypebots for Tavern: https://rentry.org/pn3hb▶ BotmakingEditors: https://agnai.chat/editorGuides: https://rentry.org/meta_botmaking_listAdvanced: https://rentry.org/AdvancedCardWritingTricks▶ /chag/ dustbin of historyhttps://rentry.org/ChagArchivePrevious thread: >>41677635▶ Current theme: Soon™**News:**- New GPT-4o snapshot gpt-4o-2024-11-20 released.- Gemini-exp-1121 and 1114 are now available via the API. Free on Openrouter too.- Pixtral large out: https://mistral.ai/news/pixtral-large/
Anchor for bots, lorebooks, scenarios.Previous >>41677636
Anchor for technical stuffs (Proxies, Updates, Models etc.)
Anchor for asking for bots, lorebooks, scenarios etc
Bots corner:RGRE Streaming: https://mlpchag.neocities.org/view?card=Anonymous/RGRE%20Streaming.pngDot Matrix: https://mlpchag.neocities.org/view?card=Anonymous/Dot%20Matrix.pngTrixie Lulamoon: https://mlpchag.neocities.org/view?card=anonistrator/Trixie%20Lulamoon.pngRolling Thunder: https://mlpchag.neocities.org/view?card=anonistrator/Rolling%20Thunder.pngInky Rose: https://mlpchag.neocities.org/view?card=Anonymous/Inky%20Rose.pngMarezilla: https://mlpchag.neocities.org/view?card=littlewitcher/Marezilla.pngPonyville Cooking Show: https://mlpchag.neocities.org/view?card=Anonymous/Ponyville%20Cooking%20Show.pngRarity EQG Update: https://mlpchag.neocities.org/view?card=Maverick074/Rarity.pngApplejack EQG: https://mlpchag.neocities.org/view?card=Maverick074/Applejack.pngEarth 2045: https://mlpchag.neocities.org/view?card=MaudPie/Earth%202045.pngScreenshots corner:Maid Dash >>41678778Trained for this >>41679403Sunny demands sex >>41679563Lick the spoon >>41679689Sanest unicorn >>41679856Midnight Spur >>41680199Therapy Session >>41680578, >>41680664Cadance's Talk-Show >>41680851
Screenshots corner 2:Thirsty Mares >>41683210, >>41683258SNIFF >>41683275FIO >>41688641Theme Song >>41691775, >>41691786, >>41691961, >>41692050, >>41694645
Screenshots corner 3:Gipsy Bard >>41692149C.O.O.L >>41692329Muffet >>41697269Velvety muzzle of a foal >>41698024
Here are the results of the previous poll. A tie and generally quite close outcomes. So I'm making another quick poll to find out which theme you want first, knowing that it will be Theme 1 -> Christmas/New Year -> Theme 2.https://strawpoll.ai/poll/vote/UIyOK7cPxLFJ
>>41698722Quite a few really close results.
Updated the rentry tutorial for local hosting and models i've tested and updated new setting file for formatting and context and new default model.>ST-formatting.jsonI've also added what i've learned here about how the model works and how ram / Vram relates to the models and what you can run.Feel free to check out the new model experiments and new formatting master import file.https://rentry.org/lunarmodelexperiments
Lovemaking with AI mares!
>>41698816I've figured it out it's my fucking jailbreak and I don't even know how it's doing it
>>41698820
>>41698857Updated my jb and it works againI don't know what happened to break them so bad but it's working like a charm now
Wanted to try out another model, but Mistral is just breaking in ST every time I choose her.
>>41698699woah wife in the op 0_0
chorbo preset if anyone wants to give it a try>>>/g/103353294
>Gemini bringing hands and fingers We’re back to the 1.3 days!
>>41699529never seen fingers on Gemini. which Gemini does that? 1121? do you have humans in RP (like Anon)?
Is Hermes 3 405b Free from OpenRouter dead?What's a good free alternative?
>>41699558 >1121? Yes. Just got a "She snaps her fingers" on it. >Do you have humans in RP (like Anon)? I’m playing as Anon. I’ll just put the usual line about pony anatomy in my preset and see how it goes.
>>41699572 Looked it up, and yes, it’s not free anymore. >What’s a good free alternative? If you have good hardware, you can try local models, there’s a guide in the OP. If you have spare "burner" Gmail addresses, you can go here and get API keys for Gemini with 50 prompts a day per account: https://makersuite.google.com/app/apikey You can also read the Novice Guide until the very end.
>>41699577>I’m playing as Anonthen I think that's the case. the presence of human in RP boosts relative domains/tokens. probably an unlucky roll. [fingers] were like 3/4 most-likely token(s) and model picked them over other options. it happens. also might be caused by temperature. Gemini support [0,2], maybe try to down temp a bit to 0.9 for example? to make sure model picks most likely tokens next time?but on other hand. if model has generated>She snaps herthen what next word it should have picked? horn? sounds weird too. and what if she is a pegasus? hoof? even more weird. wrist? pastern? heel? muzzle? tail? hock? aside from "fingers" I cant think of other option that makes sense with "snaps". so model kinda wrote itself in a corner with this selection of token. if it did>She flicks herinstead then yeah it would have better chance to pick something that makes more sense for a horse
>>41699529I mostly get the "angry red lines" on my back.
I got a question, localanons.kobolodcpp works great, the model reponds very fast and well.but in ST it reponds with straight gibberish.I use the LNR system prompt/context/presets and the model just spams ####It's also really slow to respond, it reprocesses the whole context on every reply, which the default kobold site doesn't.
>>41699854yo, im the guy who made the guide, theres 2 reason why this happens, thats the context template and instruct having a mental breakdown. what context template setup do you use and what model do you use? sometimes some model dont vibe with the context template and need a different one
>>41699899I've tried both Fizzarolli-Nemo_Pony_2-merged-Q4_K_M and MS-Schisandra-22B-v0.2.Q8_0But, I don't think it's from ST anymore.The schisandra hasn't worked for me, just spams "ckck". But I haven't tested much either.The nemopony works for the first reply but then all others are just "usus".Weirdly, doing a new chat, I could do several replies straight without issues, but when it crossed the 512 token mark it started to spam again. Even when I deleted stuff and regenerated it still spammed "usus".I don't really get where the issue is, I use your kobold settings but use vulkan.
>>41698722Two hours left.
Too many fucking ai mare models>Which Mistral is best?>Which Gemini is best?>Which GPT is best?>Which Claude is best?
New video model with weights released, by far the best so farhttps://aivideo.hunyuan.tencent.com/
>>41700114 Obligatory "this is just my opinion" blablabla: >Which Mistral is best? The latest Mistral Large 2, but it suffers a lot from repetition problems if you don’t use it locally, since you don’t have access to some repetition control options. And good luck running it locally, as it’s a 123B model. The different Nemo 12B models are good for local use. >Which Gemini is best? Experimental 2024-11-21 until Gemini 2 releases. >Which GPT is best? ChatGPT-4o-latest has the best prose and is the most fun of the GPT models. They monkey-pawed it, and it’s also the most filtered for NSFW, unfortunately. Still, I’d use it over the others, which are horrible IMO. >Which Claude is best? Claude Opus has the best prose and is the most creative. It’s also very, very, very expensive and no longer available on proxies (Claude in general isn't). If you pay, Sonnet 3.5 latest is the most intelligent model and can write well too. However, it needs a good jailbreak, whereas you could get by with a very simple one for Opus.
>>41700157 I Updated to ST 1.12.8 This is the kind of shit I mean.I don't see Mistral Large 2, just a billion other models and didn't pixtral large just come out? It's in the 'news' in the OP. how does it compare?
>>41698722Locations won.
>>41698699Anchor for the Locations event!Everything goes as long as it’s on theme.Post bots, post logs, post lorebooks...End of event: 12/18
>>41699931i updated the file for master import, the new default for context template is mistal-custom-3 (the one i modified)the other 3 i use LNR settings
>>41700171 Large latest will redirect to it. I believe Pixtral is just that model with added multimodal functionality (e.g., reading images, etc.), but I might be wrong, and they could be slightly different. Another anon can correct me on that. I haven’t kept up much with the Mistral models.
Would you let your waifu risk her life saving the world without you by her side?
Gemini was released on December 6th of last year. I wouldn’t be surprised if they release the new model for the anniversary.
Has anyone tried Nemotron? I just saw this post, and it seems interesting: >>>/g/103393142
>>41700425Tulu 3 is better imo. Knows the characters better.
Chorbo was giving me in character refusals last night
>>41700190this card has been here for a while, but might fit into the locations eventhttps://mlpchag.neocities.org/view?card=Anonymous/Canterlot%20Highschool%20Sim.png
I fucking love this greeting on celestia so fucking much because it always gives me the most fun discussions and scenarios based on my autistic RTS/map game brain rot, i wish there was a fully dedicated card to this kind of concept
>>41700190 Which places might be interesting?
>>41700543OH MY GOD IM SO RACIST AGAINST GRIFFONS AND DRAGONS NOW THEY'RE SO RETARDEDI HOPE YOU HAVE SO MANY BORDER SKIRMISHES YOU BOTH GET WIPED OFF THE PLANET
>>41700501Neat. But isn't the point of the event making it so that anons get inspired to create new cards?
>>41700813 NTA, but yeah. I won’t add old cards to new events unless there’s a total update. I remember this one being made for the Simulators event. Still, there’s no harm in linking what already exists as a reminder or to showcase old cards, as long as it fits the theme, even if they won’t be added to the new banner.
>>41700199So I can just use large latest-and not have to worry about the particulars. Thank fuck, GPT was already bad enough with all the -orbos
Update to Estrus 2Card: https://files.catbox.moe/miksdf.pngBook: https://files.catbox.moe/6bho5z.json>Found a better way to work with gpt, mistral and gemini to more consistently following the rng without needing to help the Ai by nesting the commands in [ooc] brackets. The card now functions as intended, and made some small changes to the book.
>>41701159Nice.
>>41701159>If estrus card was so good why there no estrus 2?
>>41701394Do good cards have to have sequels? Should I make a Breedinquestria 2 card?
>>41701488Your best card is the massage parlor but I don't really know if you can just slap a 2 infront of that.
>>41701514Massage Parlor 2, now with more ponies.
>>41701514What is considered my worst card? Interested in never making a sequel for this one.
This model is excellent.https://huggingface.co/sophosympatheia/Evathene-v1.3I have never found a local model that captures everypony's character that well. Its also quite smart, knows pony anatomy well and knows quite a bit about the show.
>>41701587>72B parameters>150GB modeljesus, how do you even run this?you got a supercomputer or something
>>41701488I made Estrus 2 to unleash the concept with new locations and new amres pulling from fanon, g5 and and ocs since the first one was really constricted to just ponyville and rehashing ponies youd know from s1. Imo you shouldn't ask if you should make a sequel, you should make what you want and if thats a sequel then think about what you would wanna do for one
>>41700157I told your brow QwQ is better than Mistral Slop and the other ones, is claude but local.
>>417016428 bit quant is half this size. If you have 96 gb of ddr5 you can load it to ram and have like 1 tps, which is quite sad. There is the speculative decoding approach where you need 2 models with the same vocabulary and you use the smaller one to generate faster where possible, then check result with bigger model and make sure the sequence is likely according to it. Mathematically identical performance, but they don't make much of theese model pairs now.
Amazon released their own models. It’s only available on AWS for now, so I don’t think we’ll see it on a proxy. https://aws.amazon.com/ai/generative-ai/nova/understanding/ $0.80 per 1 million input tokens and $3.20 per 1 million output tokens. Benchmeme included.
>>41701834 OpenRouter said it’s coming soon. It would be great if they aped Anthropic models for a fraction of the price.
>>41701642lots of ram to load the model, then lots of patience for the response
I love how tulu 3 70B writes Celestia.
>>417016422x 3090s or such at 4 bit
I could swipe this forever
x.com/OpenAI/status/186432892826725994112 days.12 livestreams.A bunch of new things, big and small.12 Days of OpenAI starts tomorrow.
>>41702280even with very generous count - Sora video-gen, o1 full release, update for Realtime, update to Whisper, update to DALL, some fancy new thing in chatGPT, better management for Enterprise clients, grounding like in Gemini, new snapshot for research and testing (aka prompt honeypot), update to structured output --and I still failed to reach 12 releases
>>41702310They are scrambling for something. Matched by a local model, outdone by a local video model...
>>41701159Updated both the card and the lorebook. Thanks anon.https://mlpchag.neocities.org/view?card=Tiffin/Estrus%20Simulator%202%20Electric%20Boogaloo.pnghttps://mlpchag.neocities.org/lorebook/G5.json
