I decided to purchase 1 month of nanogpt and it have a lot of models but all are super censored how do you get glm 5.2 or deepseek v4 to generate nsfw rp? I have not tried to hook it into sillytavern yet i wanted to try the capabilities first

I have not really tried this other than pasting some text back and forth, but wouldn't this be a good way to stop the AI from finding patterns within its own prose?

I have a feeling this might be really helpful in preventing prose from stagnating. Every single model I have tried (even fable) likes to find a pattern, be it prose structure, specific words, or characteristics.

It might be able to help with slop and LLM-isms as well. I have never seen anyone talk about this, so I wanted to ask if someone has already tried this and what the results look like.

Hello fellow SillyTavern Subreddit,

I am since the beginning of ChatGPT all in the search of the ultimate text adventure. Before that marvel of technology was released, I played DnD, Pathfinder and DSA with a small group of mine. But since they all lack time, I was left alone, and ended up using ai as my GM. I always liked this, without a real rulebook to walk through a generated world and experiencing it and playing with it with a character you created. My prompt reflects that, I dont use character cards, i dont like setting up endless lorebooks, instead, my prompt guides me through a setup phase where the ai offers me different settings, worlds, characters, companions and so to go along with.

Yeah, so this actually makes SillyTavern quite unattractive for me, since it heavily relies on character cards, lorebooks, worldbooks, companioncards... in addition to that the UI is pretty... eh.

Now after almost 4 years of chatgpt/functional llm's, I would say, I reached the peak what to achieve with prompting alone, but, after countless of hours of testing and playing with it. The same end of the line is reached. Long context and consistency.

What brings you the best prompt ever, if after 100k tokens context, the api costs are too high to play with, even if you are using piss cheap ai's like DS4f/p, and the consistency of the gameplay, the knowledge of the npc's the feeling to play, just degenerate faster than the True Creator can spell their honorific name.

So, now to my request, since if there is a community full of people resolving about how to goon and enjoy rp for long term, then its you guys.

I would appreciate your ways or ideas or even apps I dont know of, where this is a fixed issue, where you can achieve lots of fun with your own prompt.

To the things I already tried, in the hope they do achieve this, Marinara Engine, Old Gregs Tavern, Friends & Fables, **NOT** SillyTavern, Even claude code. With this marvel of technology, we call LLM, I also tried to build an app that could achieve this, without any success yet.

So, I am all ears, and also fully fine if noone answers to me.

Cheers.

This is our weekly megathread for discussions about models and API services.

All non-specifically technical discussions about API/models not posted to this thread will be deleted. No more "What's the best model?" threads.

^((This isn't a free-for-all to advertise services you own or work for in every single megathread, we may allow announcements for new services every now and then provided they are legitimate and not overly promoted, but don't be surprised if ads are removed.))

How to Use This Megathread

Below this post, you’ll find top-level comments for each category:

• MODELS: ≥ 70B – For discussion of models with 70B parameters or more.
• MODELS: 32B to 70B – For discussion of models in the 32B to 70B parameter range.
• MODELS: 16B to 32B – For discussion of models in the 16B to 32B parameter range.
• MODELS: 8B to 16B – For discussion of models in the 8B to 16B parameter range.
• MODELS: < 8B – For discussion of smaller models under 8B parameters.
• APIs – For any discussion about API services for models (pricing, performance, access, etc.).
• MISC DISCUSSION – For anything else related to models/APIs that doesn’t fit the above sections.

Please reply to the relevant section below with your questions, experiences, or recommendations!
This keeps discussion organized and helps others find information faster.

Have at it!

It's not an issue of Nvidia Nim, GLM 5.2 works ok through NIM and openwebui, but Silly Tavern throws a "internal server error" without any more explaining

Hey folks! We are a family couple and both are gamers. Since we discovered SillyTavern and text-based RPG, we almost stopped playing video games.

Don't get me wrong, we still occasionally hop into online games for a raw adrenaline rush. Risking high-value assets in EVE Online still gets our blood pumping and keeps the thrill alive. Yet, regular offline titles just can't hold our attention anymore. Text-based RPGs have completely rewired what we look for in a story. The empathy we feel for our own characters and the NPCs creates a deep emotional connection that standard single-player titles simply cannot match.

We wonder if this is a common pipeline for gamers turning to AI. Did SillyTavern break your ability to enjoy normal video games, or are we just way too deep down the rabbit hole?

Has someone managed to correctly import charx cards to Silly Tavern? I mean with all the functions and interactive UI aspects that they seem to have. I tried using a Risu charx extention someone else made some months ago but it seems to not work. Any new extention for it?

ive been using Freaky frankenstein 4.0 but im open to any other presets, idm the token usage etc

Hello friends, GLM (regardless of the version) repeats sentences a lot. It both rewrites what I say to understand it, and when it likes a sentence or a paragraph, it continues the story by writing almost the same word in every message. For example, in a 10-paragraph text, it writes the middle 3 paragraphs with almost the same logic. I'm really tired of this. There isn't a preset I haven't tried — Frankenstein, Chatfill, or whatever. I've tried them all. Is there a solution

I have been messing with fable for the past couple of days, and I must say it's not the best model for RP, yet I'm obsessed with breaking it. I have tried many things, playing with the prompt, settings, api, etc. And yet I can't make it write everything (not talking about coding here). It can write smut to some degree, incest, slavery, and non-consensual stuff are a bit rough to get and it absolutely does not fuck with anything with minors in it. Best I have gotten was using spiritual spell jb which allowed for incest and the others to some degree but still refuses a lot of the time.

Have any of you guys gotten better results?

(If you're going to mumble about "why would you want to disable it?", etc., just close the tab.)

Ideally, I want to strictly disable the reasoning (on the models that allow it).
&gt; "reasoning": {"enabled": false}
&gt; "thinking": {"type": "disabled"}
Both of these official parameters kinda work, but it still triggers reasoning sometimes.

Now, is there a way to manually strip it out in the SillyTavern?
So far, I have tried:
- adding ["<think>", "</think>"] in the Custom Stopping Strings. Didn't work.
- adding <think>, </think> in the Auto-swipe. Didn't work.

So I moved to local hosting after Chub threw my subscription in the trash. I am slowly but surely getting used to silly tavern/koboldcpp.

Anyway I wanted to ask if there's an ai model anyone might suggest to get me as close to an experience as chub mercury? I've been using this 12b model but i don't know, it feels a little stale for lack of a better word. I think with my pcI should be able to handle a more advanced model, I'm just hoping for something close to what chub mercury was.

If it helps, my pc specs are:

RX 9060 XT 16gb vram

I7-12700KF

32gb ddr5 ram

I use ST on both mobile and pc, where the pc is both the server and client. As I want my sessions to be stored there but be able to continue on my phone, I have remote connection enabled within my local network.

No problems so far, but both Firefox and Chrome on my Android phone will often stop processing the response when the browser app isn't on screen. This is kind of frustrating because it basically keeps me from multitasking on my phone as I wait for prompt processing.

Is there any way to work around this? Anyone have a more positive experience with this?

Hey. I wanted to know if opus is still recommended as ‘good’ in creative writing. I wanted to use it but I’m using sonnet 4.6 since 5 is pretty bad. The writing is not well and has filter.

Should I use glm 5.2 or bite the bullet and pay for opus?

My main issue with LLM roleplay is that the world constantly revolves around the {{user}}. NPCs seem to wait for my every move and are afraid to challenge or hurt me.

A while back, I read about an unorthodox RP method: the {{user}} card is kept nearly empty and functions as a narrator. The actual character is then described in a Lorebook entry instead. That way, the LLM (probably) treats this character as just another NPC.

However, if I write something like "{{user}} acts as Greg" (for example, Greg is my actual character) in the persona card, won't that just make the LLM think "{{user}} = Greg"? I don't understand how it should work.

To those who actually use this method: could you share examples of your persona cards and the Lorebook entries you use for your actual character? Also, what changes did you make to the character cards (i.e. {{char}}) to make this work?

PS: Sorry for any English