DAIsy is a Discord bot with AI Integrations
Environment Variables:
DISCORD_TOKEN=XXXXXX
IS_GPU=<yes/no>
REFINE_IMAGE=<yes/no>
Install Requirements (Conda):
conda install -c conda-forge diffusers
pip install git+https://github.com/huggingface/transformers
pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
pip install transformers[sentencepiece]
pip install optimum
pip install auto-gptq
Recommended Base Models
- Models should be cloned into a '/models' directory in the root of this repository
Chat: https://huggingface.co/TinyLlama/TinyLlama-1.1B-Chat-v1.0
Image Generation: https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0
Image Refinement: https://huggingface.co/stabilityai/stable-diffusion-xl-refiner-1.0
GPU Support
For faster inferences It is recommended to run on a device with a compatible GPU.
- PyTorch version must be compiled with cuda support
- ENV variable
IS_GPU
must be set toyes
Image Refinement
Generated images can be run through a refiner model to remove image artifacts and provide an overall better quality image.
- Image refinement adds additional delay to image generation
- To enable, ENV variable
REFINE_IMAGE
must be set toyes
Discord Commands:
!chat <text generation prompt>
!image <image generation prompt>