this project is forked from TabbyML/tabby
modified by zkytech
- For users from china and those who want to deploy this project in offline environment: remove the code to download model from network, direct load from local.
- Package model files into docker image.
- Change entry point command to
/opt/tabby/bin/tabby serve --model TabbyML/${MODEL_NAME} --device cuda \$1
- GPU is required, no CPU support because of bad experience.
docker run -it --gpus all -p 8080:8080 --model zhangkunyuan/tabby:StarCoder-1B-latest --device cuda
services:
tabby:
image: zhangkunyuan/tabby:StarCoder-1B-latest
restart: always
ports:
- 8080:8080
deploy:
resources:
reservations:
devices:
- driver: nvidia
count: 1
capabilities: [gpu]