quick hack to get tavernai to work with spellbook.scale.com for playing around with GPT4. based off of cncanon1's existing OpenAI mod.
Now supports full GPT-4 8k context size. Context size is limited to 7750 tokens in TavernAI to allow space for the system prompt.
If you don't want to set up any of the shit below, you can just download it and run. It is preconfigured to connect to a bot. Just launch Start.bat
to get started. Node.js v18 is required.
If you want your own private (lol) bot, follow the guide below.
- Create scale spellbook account at https://spellbook.scale.com.
- Create an "App" (name/desc doesn't matter)
- Create a "Variant", which sets the parameters (system prompt, model, temperature, stop sequence, respnse token limit, etc) for your bot
- Make sure to select GPT-4, you retard
- Replace the contents of the "User" section of the prompt with the following:
Complete the next response in this fictional roleplay chat.
{{ input }}
- 💡 Set the model parameters as desired.
- Model: GPT-4
- Temperature: ~0.6 - 0.9
- Maximum Tokens: 400 - 4096
- This controls the max length of the AI's responses, NOT the context size. If you set this to a very high value, you MUST adjust the "Max Response Size" value in TavernAI's settings to match!
- I recommend 300 - 500 for chat purposes. You don't want the AI sending you 4000 token chat messages.
- Click "Save New Variant"
- Go to your new Variant and click Deploy
- This will create an API key and URL for your bot.
- Launch the server using the command line by running
.\Start.bat
. - Open the TavernAI sidebar, switch the API to "Scale", and paste your API key and URL in the boxes. Click "Connect".
- Now start having fun with GPT4 :)