Ren'Py Influencing [v0.1.17.8] [Golden Crow]
- Thread starter BaasB
- Start date
-
- Tags
- 2dcg adventure anal sex animated big ass big tits character creation cheating corruption dating sim exhibitionism female protagonist footjob futa/trans futa/trans protagonist gay harem interracial lesbian male protagonist management masturbation mobile game multiple protagonist netorare oral sex prostitution romance rpg sandbox simulator vaginal sex voyeurism
Question for AI connoisseurs: is it possible to make the AI give an answer with translation or duplicate it in the browser in which kobold is open?
- Sep 14, 2020
- 79
- 183
You must be registered to see the links
(Same post as .6, just updated links)You don't have permission to view the spoiler content.
Log in or register now.
- Dec 21, 2016
- 183
- 317
Ollama is a pain in the ass and can't use quant files like GGUFs direct from huggingface (you have to create what are called modelfiles from them, tutorials abound for this stupid requirement, or pull the models directly from ollama's library). KoboldCPP is a much superior option, in my opinion.
- Dec 21, 2016
- 183
- 317
Q8s are overkill. Generally Q6 is considered near perfect. I only do Q8 when I have VRAM to waste (I have 24GB so I use Q8s for 12Bs.)
Don't be afraid of quants down to, say, Q4 with a 8-12B, but lower than that and it's going to get noticeable. I use Q4_K_S IIRC for 32Bs which can just fit in 24GB, though the larger the parameter count (B = billions of parameters) of the model, typically the better it can handle quantization.
- Jul 5, 2017
- 327
- 410
I tried Kobold but it has a similar problem I'm running into right now, can't figure out how to save models somewhere else since it installs on C directory and mine doesn't have that much memory left. Trying to figure out how to change directories atm.
- Dec 21, 2016
- 183
- 317
KoboldCPP is just a single executable with no installer. You can put it anywhere you want and point it to a GGUF anywhere it's kept. Be sure you aren't using KoboldAI which is a related project, but not the one you want.
You must be registered to see the links
Last edited:
- Jul 5, 2017
- 327
- 410
No, I didn't know that, I'm very glad you mentioned it, I was using the wrong one as you have guessed. Currently testing the CPP but I think I got it right this time. Thank you for sticking around and helping people troubleshoot stuff, users like you are the real MVP.
Edit: It worked, nice and simple. Its a big laggy but acceptable.
Last edited:
- Feb 20, 2018
- 610
- 392
So to resume the game will never have non-consensual content and will always remain a carebearing dating sim, sorry for such a game the could have been a great one...
One more game developpement killed by patreon system
- Jul 30, 2020
- 2,543
- 1,859
There is one loose end, the AI initiates sex, but the NPCs in the game refuse - you get minuses in the relationship
- Jul 30, 2020
- 2,543
- 1,859
It seems that this game still has some way to go before it's fully playable, but it's already made me laugh a lot. When you ask the headmaster about the job, you can ask him about his menstruation cycle... It's just that mine was a guy. Apart from that, I don't know, maybe I've missed something, but during the ‘construction’ of the characters... Eg: I'm a guy and I'm married. I add mum, dads and a sister when I add people, they all seem to live under the same roof as my mc. Then I want to add my wife's family, so say yes to ‘add customization family’, so I create my wife's family, but I can't link them in ‘relationship’ if they don't live with us? I mean... Crap! If I want to add friends, will they also have to live under the same roof?
- Jun 21, 2017
- 775
- 1,304
I haven't played the latest version but you can kind of get around that somewhat by cheating maximum dominance between you and the person you want to leave the house, then tell them to leave, they'll then move out into their own apartment(s) with a relationship hit but you can cheat that back. I'm not sure of what happens when you have a couple leave individually, likely they would move into their own separate apartments. As far as friends go, apparently in this newest version you can create custom, separate households, so you could create whatever friends you want and cheat an initial relationship after character creation.
- Dec 21, 2016
- 183
- 317
First thing to do is make sure the model and context are fitting into your VRAM. If you're using windows you can just Task Manager->Performance->GPU. When you load the model in KCPP you'll see your dedicated GPU memory usage spike. Make sure it doesn't fill enough to start using Shared GPU Memory. If it is, you need to use a smaller model or quant. You can also play with the layers value. If you absolutely cannot fit it in VRAM, you can split layers between GPU and CPU. This is still slow, but not as slow as it using shared memory.
The problem I see is, at the moment, Influencing doesn't seem to handle streaming from the backend. I get about 13 T/s (tokens per second) which is faster than I read so it would be fine if it streamed (displayed tokens/words as they are sent from the backend), but instead it waits for the full response before displaying it. It's possible it's because the dev wants the game to do some post-processing on it or something but it's a pretty big problem when it's too slow for even someone with a decent GPU to tolerate. Waiting 20 seconds or so for a full response is just too long for me to want to bother.
Even people who only get 5 T/s won't find it too unbearable if it streams. Hopefully this isn't a limitation of Ren'Py because it's pretty serious.
- Nov 14, 2017
- 11
- 6
I've been using this ever since I got back to the AI scene via SillyTavern:
If you ever need a way to find compatibility with your current GPU VRAM available for it, using that link may help a bit to finding compatible models for your GPU.
Running EXL2 from the local text generation has been superb, fast, and quite logical.
You must be registered to see the links
If you ever need a way to find compatibility with your current GPU VRAM available for it, using that link may help a bit to finding compatible models for your GPU.
Running EXL2 from the local text generation has been superb, fast, and quite logical.
- Sep 29, 2017
- 902
- 1,450