spara / localstack_ai

Containerized environment for experimenting with open source foundation models

Localstack AI

Localstack AI is a containerized environment for experimenting and building AI applications. It supports running open source foundation models in Ollama. Included are a minimal Jupyter notebook and PostgreSQL with pgvector.

Also included are Python packages for OpenAI and AWS Bedrock.

Pre-requisites

• Docker
• Docker compose

Optional: If you want to build pgvector

There isn't an official image for pgvector. You can build it from source. Clone the repository.

$ cd ./pgvector
$ git clone https://github.com/pgvector/pgvector/tree/master

Start a stack

Localstack AI supports Ollama and Llamfile and creates a separate stack for each. You will need to download an LLM for each stack respectively.

Download models for Ollama.

Download models in GGUF format for Lllamafile from Huggingface.

To start a stack with built images:

$ docker compose -f ollama_stack.yml up

or

$ docker compose -f llamafile_stack.yml up

To start the Ollama stack and build PostgreSQL with pgvector:

$ docker compose -f buildstack-pgvector-build.yml up

To stop the stack

To stop the stack:

$ docker compose -f buildstack.yml down

To stop the stack and build PostgreSQL with pgvector:

$ docker compose -f buildstack-pgvector-build.yml down

Running Jupyter Notebook

Docker compose outputs the messages to stdout. It will post the link and token to Jupyter notebook. Copy the appropriate link and paste into a browser. For example:

jupyter-1  |     To access the server, open this file in a browser:
jupyter-1  |         file:///home/jovyan/.local/share/jupyter/runtime/jpserver-6-open.html
jupyter-1  |     Or copy and paste one of these URLs:
jupyter-1  |         http://9faab81c039c:8888/lab?token=4d0ce25e42fd2211f4aa0b68536ff5b95b15145053d81b80
jupyter-1  |         http://127.0.0.1:8888/lab?token=4d0ce25e42fd2211f4aa0b68536ff5b95b15145053d81b80

Examples

• Ollama getting started
• Llamafile getting started