Shrimple guide into JanitorAI/Chub Venus & KoboldAI
Welcome to this simple guide on how to set up KoboldAI for free with Google Colab and connect it with JanitorAI and Chub Venus that let you have NSFW chats and more that are not allowed on Character.ai.
Requirements
You need 3 things
- A Chub Venus or JanitorAI account (JanitorAI has relatively more characters)
- A browser
- A Google account
JanitorAI and Chub Venus have rules! But unless you are a creep, you are safe. The stuff you can't do is:
- Child pornography
- Sexualized depictions of minors
- Heavy gore
- Bestiality
- Sexual violence
Have fun!
KoboldAI on Google Colab
Google Colab lets you run KoboldAI models for free. There are 2 options for models, the ones that run on TPU and those that run on GPU. There isn't much difference to you, it's just a different link you need to open. The list of the models can be found here. The one this guide is going to go with (and the one I recommend) is Pygmalion that runs on GPU.
- Open the right Google Colab page and set it up
There are two pages available
This is what you'll see
First, if you are on mobile, play the audio by clicking the "Tap this if you play on Mobile" to keep the tab alive and running.
Scroll down and go to this part, it's the only important one.
In version, choose "United" instead of "Official" and the provider I recommend to keep as Cloudflare. (also enable the Google Drive option if you want faster start times, it will download KoboldAI into your drive so it won't need to download itself the next time.)
There are already a few models pre-added, but the model we will set is not here, instead of selecting one of the models, paste in "PygmalionAI/pygmalion-6b"
This is what it should look like after setting it up.
click the gray play button under the arrow in the title (you should get a few warnings, just ignore them, it's because the code is not verified by Google, it's safe to run).
It will take a WHILE to load (depending on the size of the model 🫶). You'll know when it's done after you see links like this.
Since we will be using JanitorAI and Chub Venus instead of the UIs, only the first link is important, copy it and go to the next step. IMPORTANT: Keep the tab open, the moment it closes, the model will stop running!
- Start a chat with a character
Look for a character you like and start a chat with them, you'll be asked to set up the API if you didn't do it before (click the warning on top right or click ☰ and select "API Settings" if you didn't get a popup)
Choose Kobald AI in API and paste in the url, make sure to select "Pygmalion Format" if you use a model that has it (chat format of dialog between you and the character).
This is how it should look like (don't click "Check Kobold URL" it crashes for some reason)
Next, go to Generation Settings, just click "KoboldAI Default" to avoid any crashes and you can save and close it.
It's done! You are free to chat with your character!
The next you'll want to use it, you'll just need to start the model again and replace the url in API settings, you don't need to re-do all the steps :)
Troubleshooting
- I got an "NetworkError when attempting to fetch resource." error. What's wrong?
It's probably a cloudlfare issue, open the KobaldAI url in a new tab, if you see a "Error 1033", refresh the error page and wait a minute until trying again. If it keeps happening, restart the model following the steps again. It happens when you keep the model idle for a bit too long.
- I got a warning that character context is too long!
Shortly, whoever wrote the character made its context too long. "KoboldAI default" generation settings are set low to make the responses as quick as possible. KoboldAI has a token limit of 1000 and a context size of 4096 contexts, set these vaules in "Generation Settings" and try again. Warning: this could possibly slow down the model and make it much longer to wait for replies.
Written by PanGabka <3