Less IaC than it is collection of scripts. Allows you to easily:
- Download and manage Gguf models from Hugging Face
- Push models to Aleph as immutable SquashFs images
- Build and deploy Llama.Cpp on an Aleph VM, loading models from as immutable volumes
- docker OR podman
- huggingface-cli
- git
- jq
- aleph cli, available on pip!
- squashfs-tools, including mksquashfs and unsquashfs
All you need to get started is to copy .env.example
to .env
.
cp .env.example .env
The repository comes with suitable defaults for deploying a
simple Llama.Cpp server on Aleph, but you can modify these to
suit your needs. See the .env.example
file for more information.
Pull the configured model so its available locally
./bin/lib.sh pull-model
Push the configured model to Aleph as an object
./bin/lib.sh push-model
Pull a suitable runtime locally to build and deploy Llama.Cpp
./bin/lib.sh pull-runtime
And finally, push the runtime to Aleph, loading our configured model as a volume
./bin/lib.sh push-engine