blyatman wrote:DragOnDude wrote:blyatman wrote:is their any easy way to just download it?
Click the link, follow the instructions. There's a downloadable exe file there.
im reffering to the google part i have to have that running to work right? is there any way i can just download it whithout dealing with all the complexity
Yes you have to run the model in Google Colab.
If you know what you're doing you probably could download and run the model locally but you would need a pretty powerful computer to do that. It would be a lot more complex, actually. If you
really really want to, you can find it here:
https://github.com/koboldai/koboldai-client and go down to the "Install KoboldAI on your own computer" section. You'll still need to install Tavern AI. Tavern AI is the front-end.
Basically what you are doing is, you are running the AI model in Google Colab, the model once it is started will give you a URL. That URL links to the container that the model is running from. Then you run Tavern.AI on your local machine, which creates a local web server. So it's like a website running on your PC. Then you paste that URL you got from the Google Colab notebook. That links your instance of Tavern.AI to the model you are running in Google Colab.
how can i copy the api url from the list? it is my first time using this kind of program
After you start the model in the Colab notebook just wait, you're going to see a whole bunch of stuff happening. That's the script installing the model in your Google Colab instance. It takes a few minutes. Just wait until it is done doing everything.
When it's all done:
In black it says "
Setting Seed" <-- you need to wait until you see this. This means it's ready.
Below that in green:
KoboldAI has finished loading and is available at the following link for UI 1: (And a URL)KoboldAI has finished loading and is available at the following link for UI 2: (And a URL)
UI 1 is the one you want to copy into API url. That's the link to your instance of the model. It will be different every time. And yes you need to keep the browser with the Colab notebook open the whole time you are using it.
Ok, so I launched tavern.AI.exe, it opened up the main webpage I went to chat with a character but on the bottom it says "No Connection" and won't let me talk. Anyone got some tips and pointers for me on that?
In that web page: Go to "Settings". Go down to "API url". Paste the URL mentioned above into there. Click "Connect". Don't forget to click on Master Settings while you're in there and turn down the Context Size to ~1400.
Not sure bout the tokens part of it, I just run it and let it do what it needs to.
You need to lower the Context Size tokens because if you try to keep 2048 tokens in memory you will go over the free GPU memory limit on your Google Colab instance. There's a finite amount of memory available and if you use it all up the model will crash. If you use too many tokens the model might run out of memory before returning anything.
Seems like things are down right now? Like, I got it running like before, but now it's just refusing to connect? Anyone else getting this issue right now?
Google Colab might kick you off after a while and you will need to restart the model.
Is there any restrictions to this AI? Like can you get banned for NSFW or vore actions? or is this an offline/standalone client?
No. It's your own private little instance of the model. Or, at least, it's as private as any other thing running from a free Google service. And KoboldAI doesn't care. They have multiple models trained on NSFW content. Though... as far as vore goes... its' about as good at it as any other GPT-6b knock off is. Which is to say it doesn't really "get" it. So you kinda have to coax it.
And all your stuff is stored locally. Your chats and whatever characters you load from the presets or that you import. Look in the TavernAI-main\public\ folder. Characters has the character avatar images and Chats has the .json files which are the chat histories. It's saved locally so you can continue talking to the character if\when you need to restart the model.