- Jul 13, 2017
- 50
- 31
Others Multiic [v0.3.67] [Stvoler]
- Thread starter aleksandr unconditional
- Start date
- Aug 13, 2024
- 200
- 311
Correct me if I'm wrong but I'm guessing you were playing in storymode and haven't reached the poetry test yet?
If so, I'd recommend you keep going. There's a reason for all those "glitches". I don't want to give any spoilers but it's actually pretty brilliant once the pieces of the puzzle start coming together.
- Oct 25, 2017
- 39
- 72
They probably meant the nice 2D images, as far as I can tell they're not modelled on anyone in particular.
- Sep 27, 2017
- 500
- 946
You're the only one I've seen an example of a model that's been used. I tried this model Moistral-11B-v3-f16 and it works very slowly, 2-3 words per minute. Does it need a very powerful computer or what? Everything else I tried with this kobold didn't work at all. What models are used for this, how to use them, nothing is clear.
- Oct 25, 2017
- 39
- 72
That particular model will need a pretty beefy graphics card to run- I'm using it with a 3080 with 10gb vram and works fine. You could try using one of the smaller versions of Moistral-11B-v3, the smaller the data model the less resources it needs. If that isn't the issue then try lowering the context size in your kobold options. Higher context size and larger data models directly relate to response speed.
Of course this will give the bot less data to work with and make it more likely to repeat itself and forget things faster. That's the tradeoff; better ai needs better hardware.
Last edited:
You need a lot of VRAM, as the model is 20GB in size.
I only used it because the size doesn't matter to me. I use an RTX 3090 with 24GB RAM. However, the f16 version is labeled on hugginface as overkill. Version Q8_0 should be sufficient (half size). I have tested various other smaller models under 10GB, but I didn't like them.
If you have less than 10 GB VRAM, it will be difficult, because then your system's RAM will also be required and it will be get quite slow. If you have enough it will work, if not the model is not usefully usable unless you are a masochist
.Yes, you're right. I haven't lasted that long so far because I was a bit frustrated by the frequent interruptions. I don't have that much time at the moment either. But thanks for the tip. It has motivated me to try again
.
- Sep 27, 2017
- 500
- 946
I tried some models and the results are bad, 90% of what they produce is some nonsense that has nothing to do with the game.You need a lot of VRAM, as the model is 20GB in size.
I only used it because the size doesn't matter to me. I use an RTX 3090 with 24GB RAM. However, the f16 version is labeled on hugginface as overkill. Version Q8_0 should be sufficient (half size). I have tested various other smaller models under 10GB, but I didn't like them.
If you have less than 10 GB VRAM, it will be difficult, because then your system's RAM will also be required and it will be get quite slow. If you have enough it will work, if not the model is not usefully usable unless you are a masochist
Can we get some kind of a guide? Why does everything take so long to load to the next set of text or progression? Is what we type from a narrative perspective or the character we are playing? It seems like both work but it's all very clunky
When the content is AI created the time it takes to provide the result depends on the AI model you are using. Free online models tend to be slow. Local models will be as fast as your computer is.
It feels like a lot of models are not working too well. However the model L3-8B-Stheno-v3.2-Q4_K_M-imat works really well. That's the one I'd suggest to start with as it's only just under 5 GB in size too. It's not the best model out there, but it works well for the game.
- Sep 27, 2017
- 500
- 946
That's normal behaviour of KoboldCpp when the context is full. You can either reduce the character context or increase the context size in Kobold. I would recommend at least 8192. 12k if you can. I am not entirely sure why it does that but I believe it's related to the context shifting and it needs to do that to remove the old context to make room for new. Again, increase your context size in Kobold or reduce the character context in game.
- Sep 27, 2017
- 500
- 946
Thanks for the help but 12k, 16k context size makes no difference. I guess I'm done.
Thats funny game. In sandbox mode (Fiona) me, Fiona and futanari called "Ren" goes for adventure and some device with holographic interfece started flashing. Never thought I will play this type of game for story
From buying Fiona a pastry when my character took her to a coffee shop in sandbox mode, things spiraled to her deciding to make a pastry filled with cum and vaginal secretions (and cream), to eventually her opening an online store dedicated to selling those through a delivery app, with careful budgeting, ingredient purchases and brainstorming new product lines all factored in. Surreal. Next year I will ask her to do my taxes.
10/10
10/10