-
-
Notifications
You must be signed in to change notification settings - Fork 4
Getting started
By default it uses KoboldCpp, but if you're feeling adventurous you can change llm_config.yaml in the tale/ folder to another one.
-
Get and install KoboldCpp: https://github.com/LostRuins/koboldcpp/releases (There is also support for OpenAI api) Download a suitable llm model for koboldcpp. I recommend llama-2-7b-chat-codeCherryPop.ggmlv3.q5_K_M(https://huggingface.co/TokenBender/llama2-7b-chat-hf-codeCherryPop-qLoRA-merged), a 7b model which is both fast and very accurate (relatively speaking) when generating JSON.
-
Download repo, either with
git clone git@github.com:neph1/LlamaTale.gitor as a zip. Master branch should be stable. -
Run
pip install -r requirements.txt -
Start your backend, KoboldCpp or OpenAI compatible. (port 5001 by default, or change in llm_config.yaml)
-
Start the demo with
python -m stories.prancingllama.story
Optional:
- If you'd rather play in a browser, add the --web flag and connect to http://localhost:8180/tale/story
- There's also a GUI version, which is launched with --gui
- If you have a v2 character card and want to skip character creation, add
--character path_to_character - If you want to load a v2 character as a follower, type
load_character path_to_character_relative_to_story_folderin the game prompt - Check https://github.com/neph1/LlamaTale/wiki/Creating-a-character if you want to make a 'full featured' LlamaTale character.