Welcome to Chatbot AI General #81, the thread for discussing and improving AI pony chatbots.▶ MLP Botshttps://mlpchag.neocities.orgSpreadsheet (extra MLP bots, CAI bots, Jailbreaks, NAI Scenarios):https://docs.google.com/spreadsheets/d/1J7BeqJVDS51cXF8Pgm2YZaFq-Z6ykSJT!!! 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 opaque 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-TermuxRetards guide: https://rentry.org/Tavern4Retards▶ More frontends:Risu: https://risuai.xyzAgnai: https://agnai.chatRetards guide: https://rentry.org/agnai_guides▶ LocalsGetting started: https://rentry.org/lmg-spoonfeed-guide>>>/g/lmg/Museum of local pony projects: https://rentry.org/79hd9wLocals on Google Colab: https://rentry.org/ky239▶ JailbreaksMLP JB: https://rentry.org/znon7vxeMore JB and guides: https://rentry.org/jb-listingHypebots for Tavern: https://rentry.org/pn3hb▶ Character.AIhttps://old.character.ai▶ PAID OPTIONS (Ask the thread before considering these)▶ Openrouter (Free options are often available, like Llama 3 8B)https://openrouter.ai/▶ Setting up your own AWS instance (Claude, LLama, Mistral, etc.)https://rentry.org/how2aws4khanon▶ NovelAIhttps://novelai.netFor NAI Scenarios, Settings, and Presets, see the Spreadsheet, NAI Tab.Advanced Settings, Format, and Cards Description: https://rentry.org/82b9c▶ BotmakingEditors: https://desune.moe/aichared, https://agnai.chat/editorGuides: https://rentry.org/meta_botmaking_listAdvanced: https://rentry.org/AdvancedCardWritingTricksPrevious thread: >>41240501Chag Mini VN project: https://drive.google.com/drive/folders/1F1psRi-peKZ5box9KiOLTUkSq7_NFPUJ?usp=sharing, releases here: https://pastebin.com/AeGwrvd4News:-Mistral Nemo and Mistral Large 2 released https://mistral.ai/news/mistral-nemo, https://mistral.ai/news/mistral-large-2407, log: >>41263232-Llama 3.1 8B/70B/405B released: https://ai.meta.com/blog/meta-llama-3-1
Anchor for bots, lorebooks, scenarios.Previous >>41240504
Anchor for technical stuffs (Proxies, Updates, Models etc.)
Anchor for asking for bots, lorebooks, scenarios etc.
Bots corner:Friendship is Optimal >>41255071 https://mlpchag.neocities.org/view?card=Anonymous/FiO.pngQueen Umbra >>41259310 https://mlpchag.neocities.org/view?card=Anonymous/Queen%20Umbra.pngScreenshots corner:Element of Snark >>41240863Jousting against Shining >>41241193, >>41241213, >>41244110Nice name >>41246179Super Secret Surprise Sugarcube Slam-a-rama Shenaniganfest! >>41246512René Goscinny style >>41247042SHE DID THE THING! >>41248181It's hard being an under-sociable autistic... >>41248863Schizolight >>41248967Nurse-bump >>41250609Twilight Evolve? >>41250825Incredibly autistic and non-verbal >>41251273Celestia is a pony not a cougar >>41251520The >>41252325Little Shid >>41253515Slippery things >>41254016Snowpity >>41255055Snowpity 2 >>41255604Fucking nerd! >>41256638, >>41256994Flutterbutter >>41257111Gemini Test >>41257172Big pharma Umbra >>41259853Twilight Sms >>41260041Texting Rarity by error >>41260415Friendship is wargames >>41262056Interesting filly >>41262505Mistral large log >>41263232Too soon? >>41266454
Anchor for our fourth event focused on Alternate version/What if of characters.Post bots, post logs, post lorebooks...Everything goes as long as it's on theme.End of event: 07-31Submission: Queen Umbra >>41259310 https://mlpchag.neocities.org/view?card=Anonymous/Queen%20Umbra.png
MM...
>>41267402I'm sure he will refill soon. Just have some faith in him.
Did they make Sonnet more available on AWS after 3.5? Just wondering if Opus could follow the same path.
>>41267402Mistral Nemo atm is a decent alternative.
>>41267469Yeah, I'm happy with Large 2 on their API too.Are you the anon I just gave finetune links to? Good luck if so.
>>41267475Yep. Nemo is at the point where when given a little system prompt it does not fuck up quadruped anatomy and it knows a decent bit about pony already. I'm gonna see if I can't improve that by feeding it some high quality fimfiction fics.For 309031.95examples it look like about 10 hours per epoch. I see what a few epochs do.
>>41267496>>41267469Someone made a finetune of nemo based on Opus logs. Available on Colab for testing.>>>/g/101590406
>>41267575Yea, I used it. Seemed like a decent if small improvement. For now Im just gonna test feeding base nemo some raw fimfiction and see how it effects it. Not sure about trying to train instruct / some instruct based tune like that with just "completions".
>>41267575I'm testing it right now. It's a bit dumb when you're used to big models, but the prose and things it can come up with are really interesting.
>>41267592Yea but it follows instructions well and is smart enough for 99% of what I used Claude for besides coding. Also its dripping soul which the smarter / larger local models mostly lack. And it knows just enough about pony to get its hoof in the door with the correct terminology when told it is writing a MLP FIM story / to use pony anatomy. I think Nemo trained on fimfiction is exactly the fall back model we need. It is not really noticeably dumber than sonnet for RP / creative writing purposes in my testing. Just needs some more pony lore. Also 128K context.
>>41267575Did the same Twilight scenario test I did for the Mistral large 2. As mentioned, really interesting. You can see sentence structures that bigger models would rarely use.
>>41267463no, they didn't. AWS still very selective about giving access to models. seems to be they are having hardware problem. so no improvements in this field now>>41267608great writing and I see improvements over Sonnet. Twilight sounds a bit robotic here but it matches the scene (and model acknowledges it)
>>41267367a casting couch or confessionnal booth card
>>41267361Cute OP.
>>41267496A fine-tuned Nemo could also be run on Colab by everyone. Godspeed.
>>41268191>could also be run on Colab by everyone.I thought Google banned running LLMs for chats because of a kobold fiasco? Or it was just image gen?
>>41268198They did it for textgen a long time ago, but if I remember correctly, it was especially after Pygmalion 6B because it was used and shilled everywhere when CAI started messing things up. There were some collateral damages too, but they've stopped now.You can test a fine-tune of Nemo on it by following >>41267575, for instance.
>>41268198Google indeed banned instances running LLMs by looking for specific string. afaik they are looking for the "Pygmalion" string in notebooks and files and preventing them from launching. as simple as that. you can easily change that string to something else like “totallyLegitLLMexperiment” and be fine. Pygmalion developers are very much aware of this but decided not to push their luck and confront Google morebasically, what happened is this. in the not-so-distant past (about 16 months ago?), Google noticed a usage spike in Colab and identified that people were running Pyg via Kobold for chatting and roleplaying. they were initially hesitant to take any action because such usage does not strictly violate Colab ToS (it was created to provide environment for ML experiments, which is not that far away from using AI for chatting). but then it exploded into accusations like "Google is allowing deviants to fuck kids on their service" (because it ALWAYS comes down to that). so Google was forced to do something, and they banned this one string and reported that they "took down illegal activity". yet they left the door open for users to continue using the servicein real life, the consequences were minimal because users could use other models or rename the string. people up to this say host stuff on Colab without any negative sideeffects. free Colab allows you to host a 13B parameter model with approximately 6K context (+/-), which might be a good option for those who want to experiment with different models or merges. they do bitchslap you with limited time (like 3-4h per day per account) - but user can do multi-accounting under VPN if they really want
>>41268204>>41268222Huh, that makes sense, honestly I was aware of the ban so I got confused why someone was talking about using it. >exploded into accusations like "Google is allowing deviants to fuck kids on their service" (because it ALWAYS comes down to that). Yeah, tell me about it, literally every chatbot service has shitted the bed because of it, from CAI to isek.ai, even chub suffered from it despite originally being from /g/... though honestly it's pretty funny how that kind of people who complain are the ones that actually commit sexual harassment, ironic.
>>41268243meanwhile NAI was created by degenerates and worked for degenerates all this time. Too bad they consider textgen a second-class citizen now, and their model is way too dumb by today's standards.
>>41268301>meanwhile NAI was created by degenerates and worked for degenerates all this timeYeah but it's not free
>>41268321well there aren't many good things that are freeproxies are also paid by someone per token
>>41268329True, but since I'm a poorfag I can only leech of big corpo keys or beg OP for nai keys, and I would rather leech from some faceless corporation than from our lord and savior OP
>>41268338I'd recommend NAI only for imagegen if your GPU isn't powerful enough for local these days, honestly.
Mistral Large 2 is on Unreliable if anons want to test.