Dalai is a tool to run LLaMA and Alpaca locally, with a web UI and sockets support. It provides commands to install models and run a local server, and includes cross-platform setup guidance.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
Dalai is a project that runs LLaMA and Alpaca on a local computer. It is powered by llama.cpp, llama-dl CDN, and alpaca.cpp, includes a hackable web app, ships with a JavaScript API, and ships with a Socket.io API.
How it works
The project provides a local execution path via llama.cpp and related components, and exposes APIs (JavaScript) to interact with models. It can install models (llama or alpaca), start a Socket.io server, or run a local HTTP-like interface to query models. The README shows commands for installing models and serving a web UI, including both Docker compose and platform-specific steps.
Getting started
Getting started guidance is provided in multiple sections. Quickstart via Docker compose:
docker compose build
docker compose run dalai npx dalai alpaca install 7B # or a different model
docker compose up -d
This saves models in the ./models folder and the site is viewable at http://127.0.0.1:3000/.
Mac steps include installing node.js >= 18 and commands like:
npx dalai alpaca install 7B
npx dalai llama install 7B
Or multiple models:
npx dalai llama install 7B 13B
Windows, Linux sections provide similar commands for installing models and running the web UI:
npx dalai serve
Recent releases
Latest release: v0.1.0 (2023-03-13). Description: "Intro". The release note mentions bug fixing as a goal. The versioning and date are stated in the RELEASES section.
Traction
The repository has 12907 stars and 1327 forks. There are 334 open issues. These numbers are provided under the repository metadata.
Caveats
License is listed as none listed. The repository language is CSS. Creation date 2023-03-12 and last push 2024-06-18 are provided. No explicit licensing information beyond the field; note that the README references multiple external projects and model usage terms in the onboarding.